--- title: What Readers Said - Saturday AI Thoughts source: https://steadman.ai/newsletters/david/readers.html published: 2026-08-29 summary: Reader reactions and community voice from Saturday AI Thoughts. 26 letters sections and 21 community voice sections. --- # What Readers Said Reader reactions and LinkedIn community voice from [Saturday AI Thoughts](https://steadman.ai/newsletters/david/). 26 letters sections and 21 community voice sections. --- ## Edition #28: Fewer, bigger, better *29th August 2026* ### Letters Two replies this week, from a quiet postbag. Neither argued with the essay. One went after a throwaway line I put in a reply to it, and the other told me plainly what a weekly email does and does not do. #### A reader who looks after customer data for a living. Their verdict on the airline sale is in the email. What it had no room for is where the exchange went next. I had answered, half joking, that my own transcripts and emails, well processed, would give you a version of me eighty per cent as good, free, and that every firm has the material to build one. They did not argue about the percentage. They made two moves instead. The first was about consent: this is correspondence, and none of the people who wrote it were ever asked. That takes the question off what an old archive is worth, where I had left it, and onto who has the right to sell it. The second landed harder. Whatever gets rebuilt out of the pile, it is not the person you would want to spend an hour with. Being fun to be around just went up in value. #### A reader whose team turns insight into recommendations. The email has the forward-looking half of this reply. The half I left out is the one I have been chewing on since. They read every week, they said, and they are not sure they act differently after any of them. More often than not something in there nudges them somewhere new, and that is the whole of it. I am glad they put it that plainly. I write this wanting people to go and do something on Monday, and I quietly measure it that way, which is a poor measure. A weekly email is not a training course and it is not a to-do list. It turns up on a Saturday, and once in a while it moves somebody an inch. Nudged somewhere new, most weeks, is a better result than the one I had been aiming at. ### Community voice What readers posted on LinkedIn this week kept circling the same argument from different sides: the going out, the disagreement worth having in public, and who has to answer for a decision. #### Jay Ahern **[Jay Ahern](https://www.linkedin.com/in/jayahern)**, chief strategy officer at AFEM, the trade body for electronic music, had a week that ran both halves of this argument. In Zurich he moderated a panel on AI from the creator's side: how it's being used, creative control, rights, *"the questions artists are asking now"*. Days later he was on a float at Rave the Planet in Berlin, and the line he came home with was *"Thousands of people dancing together."* His conclusion is that technology and infrastructure matter most when they strengthen creativity, community and human connection. The going out is exactly what the payroll data says we're handing back. His whole industry is the part being handed back. [Read on LinkedIn →](https://www.linkedin.com/feed/update/urn:li:activity:7497618643522838528) #### Pavi Gupta **[Pavi Gupta](https://www.linkedin.com/in/pavigupta)**, a market-research leader at Infinity Growth Loop, an insight and AI practice, has been thinking about how rarely we let our own ideas be argued with in public. *"Feels mortifying to do this on stage."* He's about to do it anyway, in a staged duel, and his case is that a debate doesn't have to end in a disagreement: a good one leaves both sides seeing the merits of the other. That capacity, he reckons, is *"what differentiates us from the machines"*. Most organisations answer a shortage of togetherness with more meetings. One room where two people genuinely disagree does more than a year of them. [Read on LinkedIn →](https://www.linkedin.com/feed/update/urn:li:activity:7497438223414009856) #### Rahim Hirji **[Rahim Hirji](https://www.linkedin.com/in/rahimhirji)**, author of *SuperSkills*, has written in The European Business Review about the AI risk he thinks boards have left off the agenda. Not automation. Accountability. He asks what happens when *"a court, a regulator, or a customer asks 'who decided this?', and the honest answer is nobody did"*, and says the fix isn't better technology, it's *"making sure every consequential decision still has a human name against it"*. Worth turning that test on the hours. The three a week your team has quietly given back were decided by nobody either. Nothing signed, nothing minuted, and it's the biggest change most firms made this year. [Read on LinkedIn →](https://www.linkedin.com/feed/update/urn:li:activity:7497548544153862144) ## Edition #27: One inch *22nd August 2026* ### Letters Three replies this week, from a thin postbag. Each one went after a different part of last week's argument, and not one of them agreed with it in the way I expected. #### A reader who sees an upside. Last week I treated the wall of text as pure cost, the residue of a conversation nobody wanted to have. They read the same pile and saw a trigger. The output may well be "making us dumber", they said, and its sheer length is provoking deeper scrutiny rather than less. That inverts my causality. The slop is not only what an avoided conversation leaves behind, it is also the thing that forces one to start. What I keep turning over is the clock they put it on. Their correction arrives in the long run, through scrutiny accumulating over years, and it works whether or not anybody speaks up. Mine needs somebody to be brave on a Tuesday afternoon. Theirs is slower and less certain, and it does not depend on courage, which may be why I want it to be right. #### Someone else was taken with the rule rather than the conversation. The email carries their point about three letters doing the work. What I left out is that they had already run the experiment themselves, on a rule of their own that had nothing to do with AI, and reported the result honestly: it "mostly worked". I prefer that verdict to a clean one, because it points at what the compression is actually for. An acronym persuades nobody of anything. It survives being forgotten. You cannot recall a paragraph of guidance in the moment you need it, and you can recall three letters, which is enough to make the ask sayable out loud. That is the same job I described ### Community voice I also read what subscribers are posting on LinkedIn each week. This week three of them were circling the same thing the essay is about, from different angles. #### Dylan Jones **[Dylan Jones](https://www.linkedin.com/in/dylanpauljones)**, Managing Partner at Boldsquare and a fractional chief communications officer, wants the Always AI versus Never AI argument dropped. His rule is the sharpest version I have read: *use it for your first draft, use it for your second draft, use it for your 15th draft if you must. Just don't use it for your last draft.* He signs the post off as the Movement for Non-Binary Thinking. What he has actually located is where ownership sits. Not in whether the tool touched the work, but in whether you were the last one to. [Read on LinkedIn →](https://www.linkedin.com/feed/update/urn:li:activity:7493998455099514880) #### Colin Lewis **[Colin Lewis](https://www.linkedin.com/in/colinlewis)**, a marketer and retail media columnist at InternetRetailing.net, has been counting the advertising inside AI answers. Similarweb puts it in 26 per cent of ChatGPT responses, and two thirds of it lands after the second prompt. Nobody reports the performance separately, so advertisers are paying without knowing where they appeared. His read is that this is retail media all over again, and that *standards arise afterwards*. The spend always turns up before the measurement does. Knowing that is what stops you calling it a performance channel a year too early. [Read on LinkedIn →](https://www.linkedin.com/feed/update/urn:li:activity:7495409265147330560) #### David Johnson-Igra **[David Johnson-Igra](https://www.linkedin.com/in/david-johnson-igra)**, an executive marketer and AI strategist whose past clients include a16z and OpenAI, reads Stripe's purchase of OpenRouter as a bet on multi-model building rather than on any single lab. He describes OpenRouter as Twilio for language models. The number carrying his case: it raised at $1.3bn in May and sold for more than $7bn in August, a five-fold re-rate in 82 days. The part worth taking into a board conversation is what Stripe actually bought, which is visibility. Whoever can see which models are winning knows it before the market does. [Read on LinkedIn →](https://www.linkedin.com/feed/update/urn:li:activity:7495509957690175488) ## Edition #26: Go talk to them *15th August 2026* ### Letters Last week I wrote about the AI colleague that's been on my team since New Year, and argued that building one is management work rather than technical work. Three replies did something with it: one testing whether the shape of the claim holds, one already living it, and one telling me somebody has packaged the thing I said nobody had packaged. #### A reader who's come at it from the other direction.** They're already close to the same benefits, but they arrived by accumulating tasks rather than by designing anything from the top down. That's worth sitting with, because it's how almost everyone who has got anywhere with this has got there. One job that works, then another, then a third, until something starts to look like a system. Almost nobody starts with the handbook. My claim was that the handbook and the standing responsibilities are what turn a set of tools into a colleague, and their reply is the sharpest test of whether that difference is real or just the way I happen to describe it. One other thing I noticed. Reading a description of managing an AI, the word they reached for was "sounds fun". Not faster, not cheaper. Fun. #### One who's built their own.** Two things in their reply stayed with me longer than the question they asked me. The first is that they've been doing this deliberately ahead of where the organisation they work in has got to, quietly and on their own initiative. That's the bet the essay ended on, arriving as a fact rather than a prediction: the keenest person in a building is usually further along than the building is. The second is their own summary of where they've reached, which is that there's a long way to go between what can be done and what is being done. That gap is the whole subject of these emails, and I've never had it put to me more plainly. They also said they could ask a model how to build a colleague, but would rather have an answer from someone who had actually done it. Fair. #### A reader who thinks it's already a product.** The first reply to arrive, and the shortest. It named two existing products, said the third generation is basically those, and told me to go and try them. I've left the names out until I've used them properly, but the correction stands without them. The essay closed by saying nobody has properly packaged the third generation yet, and that may be half wrong. Plenty of people are packaging the plumbing: the memory, the scheduled jobs, the permission to act. What nobody can package is the handbook, because it's an account of how one particular person works and only that person can write it. So the honest version of my closing line is narrower than the one I wrote. The scaffolding is arriving fast. The management still isn't. A short postbag this week, and mostly warmth. Warmth is lovely. Arguments are more useful, so send those. ### Community voice I also read what subscribers are posting on LinkedIn each week, and this week a few of them were circling the same question from different sides: who is actually deciding. #### Rahim Hirji **[Rahim Hirji](https://www.linkedin.com/in/rahimhirji)** has a name for something worth watching. "I call it algorithmic drift. It's the point at which you're no longer entirely sure whether you made the decision, or whether the system made it for you." His argument is that this has been happening for a long time and AI has accelerated it by getting involved earlier in our thinking. The distinction he draws is the useful part: there is a big difference between using AI to help execute a thought you have already had, and asking AI what you should think in the first place. One augments judgement. The other can quietly replace it. Read the post -> #### Dave Norton **[Dave Norton](https://www.linkedin.com/in/davenortonphd)**, founder of experience-strategy firm The Collaboratives, thinks the industry has quietly stolen a word. "The AI industry uses the word to describe what the model receives. Human Context is what the person carries before the model is ever involved. Their situations, their modes, their life systems." His claim is that the data you already hold will not tell you what a customer is going to need next, and that the difference shows up in what an organisation ends up building: "Companies that get this right build customers. Companies that don't build dashboards." Read the post -> #### Dimitris Samouris **[Dimitris Samouris](https://www.linkedin.com/in/dimitris-samouris-57a99558)** at Junior sells AI software into professional and financial services. I should declare an interest: I am an investor in Junior. He wrote up six months of it from a poolside. The tell he uses to sort one kind of firm from another is a good one: "The most obvious tell to a vendor like us is whether the firm complains of 'tool fatigue' or not. The customers that are booking real ROI wins with AI are asking for more pitches, not less." He is blunter still about one product: he has not yet heard a single customer say they are booking wins with Copilot, and describes those purchases as decisions taken above the people using them. Read the post -> #### Sravanthi Kadali **[Sravanthi Kadali](https://www.linkedin.com/in/sravanthikadali)** at Persona, which builds AI interviewers, opened a case study with the sharpest sentence in this week's pool: "Ask Claude how to prep for a tech interview, and it'll confidently tell you what worked last year." Her point is that in fast-moving fields last year is ancient history, and that the thing a general model cannot match is first-hand human experience captured while it is still fresh. It is a vendor's argument, and it is also the clearest statement I have read of what a model is structurally bad at. Read the post -> #### Jesper Andersen **[Jesper Andersen](https://www.linkedin.com/in/jesper-andersen-quantum)** at Quantum took the same story as this week's third item and asked the question I left alone: who pays. "Personal AI agents are not restricted by manners, morality or social convention... the AI operates on a goal-and-reward basis." He argues consumer-protection regulators will say ordinary users cannot be expected to specify every single thing an agent must not do, and that the duty therefore falls to the AI companies. Then he supplies the counter-argument himself: "If you buy a car and hit someone, you are responsible, not the car company." Read the post -> ## Edition #25: My newest colleague *8th August 2026* ### Letters Last week's essay argued that the machines can smooth the road, but the organisation still has to decide where to go. It closed with a test: think of the last piece of work AI made quicker for you, and say what changed because of it. Some readers ran the test. Others went after the premise. #### Someone who took the question literally They had used AI to build a deck, and the deck was pointed at something specific: getting the people on their team to work in a different way. Two things improved at once. It came together faster than it would have alone, and they were clear the result was better than what they would have produced themselves. Then the honest part. The culture "hasn't changed yet", reported with a smiley and a note of frustration. That is the essay's argument arriving from the inside, and a cleaner statement of it than mine. The deck was never the thing holding the change up. #### The reader who pushed back on where the ceiling sits Their reply opened a half step away from AI, on analysis in general. Learning to work well this way, they wrote, looks a lot like learning the analytical craft in the first place. The hard part was never producing more. It was knowing how much is enough, and at what point pushing further stops adding value. They expect that call to still be theirs in ten years. They also took the first tip away to try: make the edits by hand, then send the machine's account of what changed back to the colleague who wrote the draft, so the standard explains itself instead of having to be written down. #### A reader unwilling to wait for the deeper transformation They agreed about where the bigger prize sits. The breakthrough comes when the process is redesigned around the capability, not from bolting the capability onto the process as it stands. But they were not willing to let that swallow the point. In the meantime, in the work that feeds decisions, the productivity on offer right now is real, and so is the time it gives back, and both should be banked rather than discounted while the redesign is still coming. They signed off with "no way but forward!". Two clocks are running here. My argument was about which gains convert into something bigger, not about whether the near-term ones count. #### One who took the week somewhere else They had just finished a university course on AI and arrived with a different worry entirely. They sent over ### Community voice I also read what readers are posting on LinkedIn each week. Three of them landed, separately, on the thing I've been circling: who owns the instruction. #### Conor McCarthy, Threshold Partners **[Conor McCarthy](https://www.linkedin.com/in/comccart)**, fractional AI lead at Threshold Partners, which helps consultancies and agencies build their own AI roadmaps, spent a year getting work back that was *"fast, well-made, not quite the thing I'd wanted"*, and assumed the tools were catching up. They weren't. *"I was speaking in tasks."* "Summarise this proposal" names an action. "Give me the three places where the promise is vaguer than the delivery" names a finish line, and came back right first time. Same length, completely different instruction. His line is *"the brief was always the work"*. What's worth sitting with is who used to absorb the difference. Experienced colleagues have been quietly repairing vague instructions for years, and it never appeared on anyone's objectives. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7490075898319200256) #### Marie Toft, EU AI Act compliance adviser **[Marie Toft](https://www.linkedin.com/in/toftmarie)**, who advises organisations on EU AI Act compliance, opened with the line most legal updates bury: *"The most expensive misunderstanding about the EU AI Act is that it's a law for AI companies."* The Act splits the world into providers, who build these systems, and deployers, who use them, and deployers is nearly everyone. By her reading there's no carve-out for AI that arrived bundled inside software you bought for something else, the Article 4 literacy duty has applied since February 2025, and it reaches contractors as well as staff. Most organisations are in scope, she says, *"without ever having decided to"*. A decision nobody took is still one somebody has to answer for. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7490350554758598657) #### Colin Lewis, InternetRetailing **[Colin Lewis](https://www.linkedin.com/in/colinlewis)**, a behavioural economist and retail-media columnist at InternetRetailing, has been reading OpenAI's job ads rather than its announcements. They point at an ad-sales organisation built on the Meta model and staffed with Meta people, and now at a publisher-facing network too: inventory setup, ad serving, yield. He isn't buying the destination. The stated targets are $2.5bn of ad revenue this year and $100bn by 2030, on which Colin is dry: Amazon's ad business doesn't do $100bn yet. Job listings turn out to be a better disclosure than a keynote. They describe what a company is actually hiring someone to be accountable for. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7490338741237178368) #### Jay Ahern, AFEM **[Jay Ahern](https://www.linkedin.com/in/jayahern)**, chief strategy officer at AFEM, the Association For Electronic Music, did the unfashionable thing with the Suno ruling and read the primary source. GEMA's own account of the Munich judgement, he notes, finds infringement under both German and US copyright law, confirms that licences are required for training as well as storage and reproduction, and holds that German courts may have jurisdiction even where the training happened in America. His question goes past the verdict: could comprehensive licensing become a competitive advantage for European AI companies? Every lab currently describing its models as *"licensed"* or *"ethical"* has just been handed a standard of proof it didn't ask for. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7489490348545581056) #### Danny Fahey, music and culture partnerships **[Danny Fahey](https://www.linkedin.com/in/dannyfahey)**, who leads music and cultural partnerships and writes Ear to the Ground, his own report on club culture, wants the gatekeepers back, on the grounds that not everything should be allowed in. He's watching the flood arrive as *"hundreds of thousands of hours of Housey Elevator music, AI Jazz House, and sped up versions"*, and calls taste *"the last bastion, the final frontier, the only thing that is gonna save us from being overwhelmed by a tidal wave of content disguised as good music"*. Content disguised as good music is a phrase worth borrowing for your own industry. Every sector is about to find out it has a version of it. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7490674239176527872) #### Bill Williams, IPTA **[Bill Williams](https://www.linkedin.com/in/billwilliams)**, chief executive of IPTA, an IT services firm working in American government contracting, put seven chief executives in a room to talk about AI and expected to spend it on proposals and agents. *"I walked out realizing every single one of these AI conversations was really a conversation about the human in the loop. Where the human sits. What the human's actually for."* The third question there is doing far more work than the first two. Where the human sits is org design, and you can draw it on a slide. What the human is actually for is a question about what the job was for in the first place. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7489038380899639296) [Read the email ↑](#email-2026-08-08) ## Edition #24: Smooth enough *1st August 2026* ### Community voice What readers were saying on LinkedIn this week. #### Piotr Bombol, Constans **[Piotr Bombol](https://www.linkedin.com/in/piotrbombol)**, who advises chief marketing officers on AI at Constans, spent months assembling the evidence on AI-made advertising: 17 slides, 35 studies, four field experiments. The loudest claim in the debate, that consumers hate AI ads, turns out to rest on *"plenty of polls about feelings, nothing showing measured damage"*. The finding worth sitting with: unlabelled, fully-AI ads beat human-made ones on click-through by up to 19%; label the same ads and effectiveness drops by as much as 31.5%. The label costs more than the AI does. From tomorrow, EU rules make that label mandatory. The gap between what people say they hate and what they measurably do just became a compliance question. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7487790707399774208) #### Yogesh Chavda, formerly P&G and Spotify **[Yogesh Chavda](https://www.linkedin.com/in/yogesh-ai-marketing)**, formerly of P&G and Spotify, has stopped finding "will AI replace research" interesting. His version: *"How does AI change what clients value enough to pay for?"* Synthesis, competitive analysis and early segmentation are moving in-house because AI made them cheap, and whether the outputs are always good enough *"isn't really the point"*: *"the economics of insights have changed"*. What stays external is the work where the business risk still buys rigour. Price, not capability, is redrawing the industry's boundary. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7488188315393990656) #### Michael Nevski, director of global insights, Visa **[Michael Nevski](https://www.linkedin.com/in/michaelnevski)**, director of global insights at Visa, has been deploying agentic AI in production and comes back with one discipline: *"validate against source, not plausibility"*. Agents are starting to research, decide and act on people's behalf, and his read is that the winners won't be the fastest movers but the ones consumers actually trust to act for them. Plausibility is the thing these systems are best at manufacturing. That's what makes it the wrong test. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7487537283533852672) #### Vanessa Xian, strategy and investment lead, Qatar Airways **[Vanessa Xian](https://www.linkedin.com/in/vanessaxian)**, a strategy and investment lead at Qatar Airways, spent an Oxford EMBA exchange week at Yale watching a finance professor *"walk us through using Claude Code to construct and analyze the tangency portfolio"*. Quantitative finance taught through a coding agent, in the classroom, as the normal way in. The tools arrived in the syllabus before most firms have put them in the workflow. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7488136932716322816) #### Dylan Jones, managing partner, Bold Square **[Dylan Jones](https://www.linkedin.com/in/dylanpauljones)**, managing partner at Bold Square, a communications advisory, watched Paramount pause its Warner Bros. Discovery deal and recognised the room: he sat through the EMI carve-up between Universal and Sony, and the Scripps wait for Discovery before that. His counsel is the standalone narrative, the story you keep *"behind glass, waiting for us to break in case of emergency"*, ready for the day the deal dies. People don't wait for information. They fill the gap themselves, and the good ones start taking calls. Nothing to do with AI, and everything to do with what happens to value while everyone waits. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7486550121556615168) [Read the email ↑](#email-2026-08-01) ## Edition #23: Burned and earned *25th July 2026* ### Letters Last week's edition shared your futures and fears, and admitted I'd stopped trying to reconcile the excitement with the dread. A smaller postbag this week, but a sharp one: the hardest question I've been asked in months, a pull sideways into the energy grid, one genuinely bright reading of the data, and a reader who has been auditing my writing for weeks and finally caught me. #### A reader thinking about the millions already struggling economically Their challenge led the email: telling people to take charge of their AI future is an impossible demand for most. The part I couldn't fit was where the argument went next. The tone of this debate, they observed, keeps getting darker. And if the people building the technology are right that the world changes fundamentally by 2030, then for someone with no spare money, time or leverage, what even is the point of the standard advice? It is the sharpest version of a question this email keeps circling: transformation guidance is written for people with options. I told readers to remove the roadblocks in their organisations. This reader wants to know what the advice is for the people who have no organisation, and no options, at all. #### A reader midway through a career change into clean energy They pulled the conversation somewhere no other reply went: how AI and the energy transition pull against each other, two booming sectors with what they called a "love hate" relationship. Data centres are driving up electricity demand, often met by gas turbines. Yet AI is also the most plausible operator of a smarter grid, matching supply to demand. It could trim the enormous cost of connecting renewables. They noted that the projected bill for upgrading the UK's grid to carry them has ballooned by half, to around £90bn, with a live debate over how much a smarter system could claw back. Along the way they have been using AI to learn the energy market from scratch, which is its own quiet case study in the technology they're describing. Whether AI ends up the grid's biggest burden or its best operator is a future worth watching. #### A reader deep in the numbers The brightest note in the email came from this reader: watching the data from their own corner of the economy, they are growing steadily more positive about job creation and productivity. The second thing in their reply deserves a mention too. Their reply then turned to education, a question they want to work through properly rather than dash off in a reply. That makes two readers this week who arrived at education from opposite directions: one pessimistic about the market young people are walking into, one optimistic about the jobs the technology will create. When the bright reading and the dark reading of the same facts both land on how we prepare the next generation, that is probably where the argument actually lives. #### A reader who shares the fear and the wonder alike Their education proposal made the email. What didn't was the confession attached to it. They have spent many weeks reading these editions trying to spot which parts were written by AI, or slipped past my Check, Edit, Own pass. This week, "finally (!!)", they caught a clear example, with a photo as evidence. Fair catch, and I'm oddly pleased. Every edition is a collaboration between me and the machine, and the promise is that I check, edit and own the lot. A reader auditing that promise line by line, and needing weeks to find a slip, is the system working as intended. I replied, owned it, and explained how a last-minute edit of mine caused it. In the meantime, consider the sport open to everyone. ### Community voice I've also been reading what readers post on LinkedIn. This week they were less interested in what the models can do than in the questions sitting underneath the build-out: who's in charge, whose judgement we're trusting, and whether the whole thing actually gets delivered. #### Nigel Shardlow, independent consultant **[Nigel Shardlow](https://www.linkedin.com/in/shardlow)**, an independent consultant with a background in agent-based simulation, went looking for Thomas Hobbes's grave and came back with a question about AI. Hobbes built *Leviathan* around the state as an *"Artificiall Man"*, an agent made of people but not itself human. We authorised it, Shardlow points out, because the alternative was worse. He thinks the AI agents we're now waving through, millions of uncoordinated decisions at a time, deserve the questions Hobbes asked. *"By whose authority do they act? What are we getting in return? And is the trade worth it?"* That last one is what this week's essay keeps circling. We've got very good at authorising things before we've priced them. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7484950256409976832) #### Aakash Gandhi, partner, L.E.K. Consulting **[Aakash Gandhi](https://www.linkedin.com/in/aakashgandhi)**, a partner at L.E.K. Consulting who leads its technology and digital-infrastructure practice in Asia, has been picking apart the data-centre boom from the investor's side. Demand is real, he grants. That isn't the risk. The risk is whether a platform can turn signed contracts with the big cloud providers into working capacity. That means on time, at the promised scale and margin, when transformers, switchgear and skilled crews are all constrained. His line to hold onto: *"Contracted megawatts should not be treated as operating megawatts."* A slipped completion date compounds into deferred revenue, penalties and lost future awards. The build-out everyone quotes as inevitable still has to be poured, wired and cooled, one delivered megawatt at a time. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7485574374347960320) #### Rahim Hirji, keynote speaker and author, SuperSkills **[Rahim Hirji](https://www.linkedin.com/in/rahimhirji)**, a keynote speaker and the author of *SuperSkills*, read the AI power lists and reached for a riddle from Game of Thrones: *"Power resides where men believe it resides."* The machine, he argues, is the sellsword that will produce anything for anyone at a price per token, so the deciding belief moves to whoever chooses to act on its output. When a technology makes intelligence abundant, power shifts to whatever stays scarce. He calls the people who hold it the **judgment class**: those trusted to know which of the machine's answers to believe, and believed when they act on one. Lloyd's of London, the insurance market, is already writing policies against AI hallucination. A market is forming around a single question: can you trust what the machine just made? [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7483991341106966528) #### Piotr Bombol, Constans **[Piotr Bombol](https://www.linkedin.com/in/piotrbombol)**, who helps teams adopt AI at Constans, the consultancy he co-founded this year, pushed back on the idea that consumers hate AI ads. Three studies, from System1, Ipsos and Behavio, keep finding most people can't tell the difference, and when they do suspect it, they often like the ad more. The backlash is real, he says, but it lands on *bad* AI: the rushed, uncanny stuff. Ask marketing chiefs what is actually blocking them and it isn't a customer revolt, it's time, cost and who owns the rights. The quality bar didn't move. It just got harder to hide who cleared it. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7485594358956785665) #### Elizabeth Oates, Vice President of Consumer Insights, Molson Coors **[Elizabeth Oates](https://www.linkedin.com/in/elizabethknoxoates)**, Vice President of Consumer Insights at Molson Coors, the brewer, wrote the least AI post of the week on purpose. She qualified for the Boston Marathon. *"No AI tool could run the miles. No algorithm could push through the early mornings."* Some things stay worth doing precisely because nothing can do them for you. A fitting note to end on, in a week spent counting what the machines cost. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7484003546447650816) [Read the email ↑](#email-2026-07-25) ## Edition #22: Futures and fears *18th July 2026* ### Letters Last week I asked for your big fear, the one that keeps you up on a hot night, and you sent them in. Some were close to home, some were about the shape of the whole economy. A few of you ignored the question entirely and did something more interesting: you took the Sarah story and turned it around. #### A reader who reworked the whole scenario from the worker's side This reader didn't argue with the Sarah story so much as follow it out the other side. Start with the firm: instead of hiring a person, you buy a senior strategist or engineer off the shelf from an AI lab, arriving pre-packaged with its own tools, compute and context. The lab charges you less if you grant the agent, a company in its own right, options on the work, so everyone has skin in the game. Now flip it. A high-context person could licence their own archive, anonymise the sensitive parts, and take equity in their own distributed instance. They don't leave the firm; they become "portable infrastructure", still in the loop, but owning a slice of the thing that might replace them. Same technology as my Tuesday story, pointed the other way. #### A reader who once built a life around another company's tools Their fear was the rug-pull. You build your life around something free and indispensable, the way many of us did around Google, whose email, photos and video became the furniture of daily life. Then the terms quietly change: search traffic drains away into AI summaries, tools that were sharp go blunt, cost creeps in. With AI, they expect the same arc, plus other switches they can't yet predict, sprung once we've all become reliant. Their second point was sharper and more personal. They've started refusing to engage with anything online that reads like AI, not on any high principle about meaning over messaging, but on plain indifference: "detectable slop", they call it, and if the writer couldn't be bothered to write it, they can't be bothered to read it. #### A reader who looked past their own job to the shape of the economy This reader looked past any single job to the shape of the whole thing. Today's economy is roughly a pyramid, widest at the bottom, clearest in consulting and finance but true almost everywhere you look. If firms hand the lower rungs to agents, does that pyramid become a diamond, thin at the base, and if so, what does a career even look like inside it? Their nightmare wasn't personal but civic: that work gets so commodified we all slide into a gig economy and lose the protections earlier generations fought for, the healthcare, the weekends, the holiday. If employers do less good for their communities by shedding human roles, they asked, should governments make them offset it somehow? Big questions, and I don't think we're asking them anywhere near enough. #### A reader who feared abundance, not scarcity Most fears I was sent were about scarcity: too few jobs, too little protection. This one inverted that. The reader began with the brief window of arbitrage we're living through now, where anyone fluent with the tools looks superhuman in a meeting because nothing falls through the cracks. That window closes fast, they thought, once anyone's expertise can be turned into a capable stand-in agent, which is the Sarah scenario arriving from the senior end. But the deeper worry was philosophical. Society is bearable, they argued, only because resources and energy are limited, so we let most slights go: too much hassle to fight everything. AI has limitless patience and never sleeps. When everyone has an agent willing to contest every parking ticket and every marginal penny, we get an "incredibly conflictual society". Their consolation was thin: people our age are near the end. Their children aren't. #### A reader whose fear sat closer to home Where others feared for their firms or the economy, this reader feared for something smaller and harder to defend: attention itself. The picture they painted was of more and more of us "sedated" by a constant stream of personalised, AI-generated content, tuned not to inform, teach or inspire but simply to keep us consuming. The email carried that much. What sat underneath it, and what stayed with me, was the consequence they drew: that the cost isn't just wasted hours but a ceiling on who we become. Kept gently occupied and endlessly served, we never quite reach our potential as human beings. It's the quietest of the fears I was sent, and maybe the one with the fewest contracts or clauses to head it off. #### A reader who asked who owns what you learn on the job A related reframe, but pointed at a different question: not how the deal gets structured, but who has the right to what you learn. This reader's realisation, reading the Sarah story, was that the interesting move isn't building agents but becoming one and licensing it. Everyone's selling a book, a course, a growth path; why not licence your own thinking, distilled from everything you've done digitally over a working life? Which raises the awkward part. Who owns that footprint when you built it inside a company? They paid for your time and your knowledge, so do they own what you did there, or do you get to pour all of it into your own knowledge pot and your own agent, and build your own wealth alongside theirs? I don't think most employment contracts have an answer yet. They will need one. #### A reader who watched junior colleagues shine, then wondered A lighter note to close, though it circles the same worry as this week's exam story. Sitting in an all-hands on a subject they care about, this reader watched several fairly junior colleagues present with a polish and a storytelling authenticity that was night and day better than similar meetings a few years ago. Genuinely impressive. And yet: a lot of eyes stayed down on a prepared script. Which prompted the honest question. How much of that would survive an unplanned, on-the-fly exchange, the kind you can't rehearse with a model the night before? Maybe the practice of preparing something really well carries over into the rest of the work, a halo effect. Or maybe, as they put it, it's elephants all the way down. Either way, it's the same gap the take-home exam opened up, seen from inside a meeting. ### Community voice What readers were posting on LinkedIn this week, less about what the models can do and more about the work that goes in around them. Five voices, on where AI slop comes from, why a more capable model still rewards whoever did the prep, the advantage that comes from rewiring the operating model rather than buying a tool, AI as the memory that stops walking out of the door, and scepticism as the scarce skill once making content gets cheap. #### David Johnson-Igra, executive marketer and AI strategist **[David Johnson-Igra](https://www.linkedin.com/in/david-johnson-igra)**, an executive marketer and AI strategist, has a clean theory of where AI slop comes from. Everyone prompts Claude or ChatGPT the same way, so everyone gets the same output. *"Without rules that govern the model's outputs, which uniquely reflect your brand or approach, the model's inherent rules will treat your output the same as everyone else: aka slop."* His fix is a **second brain**, a system of rules that tells the model how to think, what he calls a context harness. The point worth keeping: the slop isn't the model's fault. It's the absence of you in the instructions. Skip the rules and you inherit the average of everyone who came before. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7482895045444968448) #### Tim Ryan, co-founder, Steadman **[Tim Ryan](https://www.linkedin.com/in/timothyryanuk)**, co-founder of Steadman, watched Claude Fable 5 arrive able to finish around 16% of remote-work projects on its own, roughly double the model before it, and then spent a week watching how little changed for the people he works with. *"The thing holding them back was never the engine. It was the prep."* He borrows a line from Nufar Gaspar: context has a shelf life of about eight weeks, and most people never lay it down at all. *"A more capable model with a thin brief is still working from a thin brief."* The upgrade rewards whoever already did the work of handing the model something to work with, and does almost nothing for whoever didn't. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7482764435305074688) #### Colin Lewis, behavioural economist, Robotenomics **[Colin Lewis](https://www.linkedin.com/in/colinlewis)**, a behavioural economist who writes the Robotenomics newsletter, pulled apart a talk P&G's Marc Pritchard gave twice this year. The quotable line is that AI now lets *"every brand-builder... be a direct-to-consumer entrepreneur"*, prototyping ads in minutes rather than weeks. But the part that actually matters is buried lower: P&G moved almost all its media planning, and some of its buying, back in-house. That's the real story. The AI dividend didn't go to whoever bought the smartest tool. It went to the company willing to rebuild its own operating model around one, hands on the keyboard. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7482450225249394688) #### Pavi Gupta, market-research leader **[Pavi Gupta](https://www.linkedin.com/in/pavigupta)**, a market-research leader behind the Infinity Growth Loop, has a name for the waste in his industry: **insights-slop**, the validatory study fielded to answer a question the organisation already paid to answer, the answer sitting in a slide deck or a departed colleague's memory. His argument is that AI's job here isn't to generate more research. It's to hold what the company already knows. *"An indexed archive covering decades doesn't leave when the person does. It compounds."* His triad lands it: *AI remembers. Insights connect. Humans decide.* The scarce asset was never another survey. It was the memory that keeps walking out of the door. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7483184858735853568) #### Vivian Mohr, Bauer Media **[Vivian Mohr](https://www.linkedin.com/in/vivian-mohr)**, at Bauer Media, the publishing and broadcasting group, shared a summer reading list, and one entry reads like a note to the whole industry. On *Calling Bullshit*, he writes: *"As AI makes creating vast amounts of content easier than ever, the ability to question assumptions, spot weak arguments, and separate signal from noise becomes an increasingly valuable professional skill."* When making is nearly free, the job quietly moves to judging. His shelf is really a scepticism toolkit for an age of infinite content. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7481255767438815232) [Read the email ↑](#email-2026-07-18) ## Edition #21: Sarah stays *11th July 2026* ### Letters A thin but thoughtful postbag this week. Most of it was warm agreement with the tennis piece; the substance came from three readers who each took the essay somewhere I had not, and two of whom wrote back a second time to push further. #### A reader who spends their days moving between framing a problem and reviewing what comes back Their first move was to correct me: I had written as if reviewing were the muscle we now need, and they argued framing and reviewing are one iterative act, not two, because each review reshapes the question and not just the answer. Their second move was quieter and, I think, more useful. Working alongside a model has made their days more intense, not calmer. The gap between asking and getting an answer is a few idle minutes, and the temptation is to fill it by setting a second and third run going, until you are diving in and out of several hard problems at once and doing none of them well. Their fix was almost old-fashioned: block time to stay with one thing, what they called "the return of the Tomato principle". The muscle that has gone slack, they wrote, is the willingness to sit in the framing-and-review loop at all. #### A reader whose product is really the team they build around a hard problem In the email I gave their reframe: reviewing was never only a senior check, it verified you were building real human capability, and heavy AI assistance quietly weakens that signal. The second move was to ask what that means for value. Their hypothesis is that the product itself has not changed. People still want people alongside them in difficult moments. What has changed is the verification of the product's worth. If the consultant is no longer the author of the model or the owner of the pen, the firm needs a fresh account of what the human-and-AI combination adds beyond the machine working alone, and a way for the client to feel confidence in it. Their sharpest question folded back on my essay: is AI being used to reach the minimum information a decision actually needs, or is it just adding volume? If it is volume, the review has become more laborious without becoming more valuable, which is exactly the malaise I described, seen from the other end. #### One reader They wrote to own up that the tip about asking a model to build an HTML slide deck instead of PowerPoint had been, in their words, "my best kept secret", and to concede I was right that it was worth sharing. What stayed with me was the second half of their note, on how they read these emails. They keep falling behind, they said, not from lack of interest but from the opposite: they save every edition, refuse to skim, and want to digest every morsel for future inspiration. It is a small thing, but in a week when three other readers wrote about doing everything faster, someone deliberately slowing down to get more out of the reading was the note I kept coming back to. ### Community voice What readers were posting on LinkedIn this week, drawing a line that pairs with the essay's turn towards AI's downsides: take nothing on faith. Four voices, on understanding the research yourself, wiring the tool into the business, scrutinising the power gathering around a few founders, and keeping the freedom to switch models. #### Sameer Modha, Measurement Innovation Lead, ITV **[Sameer Modha](https://www.linkedin.com/in/sameermodha)**, Measurement Innovation Lead at ITV, the British commercial broadcaster, has a study habit worth stealing. Take a piece of frontier AI research you can't follow, feed it to your favourite model, *"whack the thinking level up to Max"*, and ask it to explain what's going on and why it matters. If you're feeling flush, point it at the code too. *"It's ok... I'll wait."* The best explainer of what these models do is now the model itself, run at full power. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7479984417999036416) #### Nick Graham, founder, Vertemis **[Nick Graham](https://www.linkedin.com/in/npgraham)**, founder of Vertemis, a research and analytics consultancy, wants to shift the conversation: *"Everyone is talking about AI tools... but not enough people are talking about AI systems."* His argument is that most AI initiatives don't fail on the technology. They fail because the tool sits to one side of the business instead of being wired into how the work actually happens. The disappointment people pin on the model is usually a plumbing problem. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7478759324467593217) #### Jesse Kanner, founder, The Write Kit **[Jesse Kanner](https://www.linkedin.com/in/jessekanner)**, founder of The Write Kit, a content and writing studio, sat with a discomfort worth naming. The New Yorker's profile of Sam Altman, the chief executive of OpenAI, struck him as *"pretty creepy and over the line"*, and at the same time as *"extremely precise and careful... very well sourced with ample skepticism applied"*. He calls Ronan Farrow, who worked on the piece, *"one of the great journalists of our era"*. Both things hold at once. As AI power gathers around a few founders, that scrutiny is the price of accountability, and worth the discomfort. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7479854579443191808) #### Phil Leslie, Chief Technology and Innovation Officer, Cornerstone Research **[Phil Leslie](https://www.linkedin.com/in/phil-leslie)**, Chief Technology and Innovation Officer at Cornerstone Research, the litigation consulting firm, read the news that Anthropic had overtaken OpenAI in enterprise share and called the horse-race framing *"the wrong lens"*. When the technology improves this fast, the leaderboard is *"a snapshot of a moving object"*. So the value, he argues, *"is not in the model you choose today, but in your capacity to switch when the frontier moves"*. Buy optionality. This quarter's best model is next quarter's switching cost. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7481027200150900739) [Read the email ↑](#email-2026-07-11) ## Edition #20: Look at the grass *4th July 2026* ### Letters A thin postbag this week, and I would rather run it short than pad it. Most of the replies to "A thousand small bargains" were kind one-liners or notes on other matters. One reader, though, sent the reflection the essay was hoping to provoke, so this week the section is theirs alone. #### A reader who thinks a lot about what makes an output good The essay's move was to hone your judgement about what truly matters against what is merely your own preference. This reader has been circling the same line for years, under a word I liked: "handwriting". They rate an output highly, they wrote, but cannot always tell whether it is genuinely better or just closer to how they would have done it themselves. Decades of client work make the two correlate, but not cleanly, and style, wording, graph formats, slide density, is where the doubt lives. Their own preference might even cost the client, yet be worth keeping as part of a product that is good on balance. The only way to know, they argued, is to run the test: leave a bit out, ship the team's version less edited, and see what happens. Doing that has made their work better and themselves more open-minded, at the price of more wobbles mid-project, all recoverable. ### Community voice What readers were posting on LinkedIn this week, mostly drawing the same line the essay does: which part of the job stays yours as the machine takes over the producing. Three voices, on the jobs data still being too early to read, on enterprise buyers keeping the routing decision for themselves, and on the seat time that builds the skills. #### Phil Leslie, Chief Technology and Innovation Officer, Cornerstone Research **[Phil Leslie](https://www.linkedin.com/in/phil-leslie)**, Chief Technology and Innovation Officer at Cornerstone Research, the litigation consulting firm, wrote the week's calmest pair of posts. On jobs: economic research hates a non-result, and the AI-and-employment debate wants a verdict either way, but "we're getting a non-result because the treatment has barely been administered". Chatbots helped without moving the needle, and scaled agent-based AI is only starting. On money: roughly 80% of Cornerstone's AI spend comes from 10% of its users, and when costs rise his first instinct is not an audit of the power users but "to learn from them", because the high costs almost always correlate with the firm's most valuable work. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7478504186305884160) . [and the costs post ->](https://www.linkedin.com/feed/update/urn:li:activity:7477810751538388992) #### Aakash Gandhi, partner, L.E.K. Consulting **[Aakash Gandhi](https://www.linkedin.com/in/aakashgandhi)**, a partner at L.E.K. Consulting, the strategy firm, keeps hearing enterprise AI buyers ask a different question. Not "which model is best?" but "how much dependency should we allow any one provider?" He remembers the cloud lesson: lock-in got painful once the workloads were deep in. So buyers now treat model choice as an architecture decision, routing each job to whichever model fits rather than wiring everything to one. His close is the one to sit with: *"Frontier labs may be winning the benchmark race. But they may also be training enterprise buyers not to depend on them."* The model does the producing. The routing is the shot these buyers want to keep framing themselves. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7477601835139366913) #### Deanna Adams, director of integrated marketing, Qualtrics **[Deanna Adams](https://www.linkedin.com/in/adamsdb)**, director of integrated marketing at Qualtrics, the experience-management software firm, borrows a phrase from motorsport: *"seat time."* The hands-on hours are what let you carry more speed out of the turns. She means marketing. But the AI skills everyone's meant to have now work the same way. Nobody frames a prompt or reviews a draft well on day one. That's seat time too. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7477345427122442240?commentUrn=urn%3Ali%3Acomment%3A%28activity%3A7477345427122442240%2C7477402369186746368%29&dashCommentUrn=urn%3Ali%3Afsd_comment%3A%287477402369186746368%2Curn%3Ali%3Aactivity%3A7477345427122442240%29) [Read the email ↑](#email-2026-07-04) ## Edition #19: A thousand small bargains *27th June 2026* ### Letters A short postbag this week, but a usefully argumentative one. The replies pulled hardest on two threads from "Average by default": who actually loses when AI personalisation drifts to the middle, and whether the work people quietly do every day is even legible enough to hand over to a model. A few readers extended the argument with their own experiments; one or two pushed back on the premise. #### A reader who works with founder-led businesses on AI adoption They came back the next day with a small experiment to run alongside the year with ### Community voice What readers were posting on LinkedIn this week, mostly drawing the same line the essay does: most AI handovers are good bargains, and the part that stays yours climbs a level. Six voices, on heritage the model can't copy, the response that matters more than the tool, the cultural two-thirds of AI's impact, the engineering that stopped mattering, trust over visibility, and choosing as the scarce thing. #### Rahim Hirji, author of SuperSkills **[Rahim Hirji](https://www.linkedin.com/in/rahimhirji)**, author of *SuperSkills*, a book on staying human as the machines get cleverer, wrote about a shortbread biscuit his family calls **nankhatai**: evolved from Persia through India and East Africa, received without thinking until the day it stops. His line on the machines: they can copy the surface of almost anything, *"but not the hands that made it, or the reason anyone bothered."* The essay this week argues with him, generously. The hands and the reason are exactly the part that stays yours in a good bargain. The trade isn't heritage for convenience. It's deciding which throughline you keep making by hand. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7474718668640378880) #### Norma Garcia, CEO, NRJ Media Group **[Norma Garcia](https://www.linkedin.com/in/normagarcia-ai)**, CEO of NRJ Media Group, a multimedia storytelling venture she co-founded, gave a CineEurope 2026 keynote in Barcelona, "Every Great Story Begins with the Unknown," that walked the room from fear to experimentation. Her thesis was four words: *"AI isn't the story. Our response to it is."* The tools will keep changing; what matters is the judgement brought to them. That's the part of the handover that doesn't transfer. You can give the model the task. You can't give it the call on whether the output is good enough, or whether you'll own it. Those climb a level and stay human. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7475503940881559552) #### Christina Habib, Chief Insights Officer for Beauty and Wellbeing, Unilever **[Christina Habib](https://www.linkedin.com/in/christina-habib)**, Chief Insights Officer for Beauty and Wellbeing at Unilever, the consumer goods group, brought the number that explains why so many AI rollouts stall: 67% of AI's impact comes from cultural factors, not infrastructure competence. She calls the rest a structural shift, not a continuum. The figure reframes the whole adoption question. If two thirds of the value sits in how people work rather than which model they bought, then the bargain that matters isn't the one with the vendor. It's the thousand small ones each person strikes with their own job. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7474499142627598336) #### Will Chamberlain, director, L.E.K. Consulting **[Will Chamberlain](https://www.linkedin.com/in/williamjchamberlain)**, a director at L.E.K. Consulting, the strategy firm, took his first paid Waymo and found the driving was the boring part. The AI handled four-way stops better than he would; he read his emails. What pulled riders back wasn't the self-driving wizardry. It was that the aircon was on and Spotify synced. *"Adoption may depend as much on experience as on engineering."* Once the hard technical thing becomes reliable, it stops being where the advantage lives. The decision that matters moves up to the part the engineering can't settle: whether the whole thing is any good to sit inside. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7475087803685515264) #### Yogesh Chavda, marketing strategist **[Yogesh Chavda](https://www.linkedin.com/in/yogesh-ai-marketing)**, a marketing strategist and ex-P&G and Spotify brand builder, watched everyone ask the same question about AI shopping: *"How do I get my brand recommended by ChatGPT?"* He thinks that's the wrong question. 64% of consumers now use AI to discover products, but consumers act when they feel confident, not when information exists. *"The biggest challenge in AI commerce isn't recommendation, it's trust."* Visibility is the part a model can hand you. Trust is the part it can't. People will let AI pick their paper towels and not their infant nutrition, and knowing which is which is human work that just got more valuable. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7474781996968325120) #### Searsha Sadek, founder and chief product officer, Shimmr AI **[Searsha Sadek](https://www.linkedin.com/in/searsha-sadek)**, founder and chief product officer at Shimmr AI, a book-advertising company, did the arithmetic on reading. Six hundred books in a life, one a month for fifty years, against a catalogue growing by more than a million English titles a year, *"and AI is only speeding the making up."* So discoverability stops being a marketing problem and becomes a survival one. When making more is nearly free, the scarce thing flips to choosing. That's the move the essay keeps coming back to: the more the machine can produce, the more the deciding is the job. Reserve your deepest effort for the few choices where being great beats being good. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7475900052398546945) [Read the email ↑](#email-2026-06-27) ## Edition #18: Average by default *20th June 2026* ### Letters The postbag was pointed this week. Five replies argued with the essay's premise, its measurement gap, and its economics from different angles. The thread underneath: the bill is the easy bit, the bias and the judgement around it are not. #### A reader who reframed the spending question Their first argument sharpened the essay's premise: the variable to manage is not the incentive structure but the user. Comfort with spending company money, they argued, is set in childhood rather than by employer values, and it shows up everywhere from meal budgets to hotel choices. AI is the first place it has really mattered at scale. The second move was harder. They suspect the pattern is not neutral. "In groups", confident insiders, are likely to spend more freely. "Out groups", quieter or newer voices, will spend less. If that intuition is right, then a flat floor-plus-budget policy could entrench inequities rather than fix them. The cautious would stay cautious, the relaxed would stay relaxed, and the gap would widen at the very moment usage starts to compound. A policy question worth thinking about. #### A senior AI lead Their first move was the practical gap the essay left open: no decent model yet for measuring cost against benefit, partly because nobody is willing to audit usage without looking paternalistic. The second move pulled the same thread into hiring. Graduate recruitment is the live question inside their organisation. They keep hiring new talent, they are adapting onboarding and job descriptions and role expectations around AI skills, and they will be the first to say they do not yet know how to manage young people in a new way. Their instinct was a practical one: convene a small group of human-resources leaders and business leaders on exactly this topic, before the next intake. The honesty was the striking part. "We have a long way to go" is a rare sentence from a senior leader on a public question. #### An associate building a market model under a four-day deadline The email covered the headline finding, that heavy reliance on the dearest model paid for itself even with the prompting overhead. The second point is more useful for anyone trying to calibrate where the premium model earns its keep. From scratch, the dearest model is "not quite there yet". Structuring and wiring an existing model together, plugging the parts in, sequencing the logic, this reader called "genuinely impressive". The implication for the floor-plus-budget argument: the premium is not paying for the model to do the work end to end. It is paying for the model to do the connective tissue that a junior under deadline pressure has the least time to do well. That is a specific use case worth calling out. #### A reader working on how organisations are restructuring around AI They picked up the essay's loose ends rather than its central argument. Training budgets are still well below where they need to be, they argued, and the "too busy anyway" reflex is quietly baking in lower performance and unnecessary cost for years to come. The deeper question is firm shape. If a graduate intake of a hundred could now be done with twenty, the better question is not how many people you need but what they should be doing instead. Then the warning: more firms will fail as a single person leaving, because the human-resources slack that used to absorb a departure is gone. The cost of the meter and the cap, on this view, is not the spend, it is the absence of a thought about the organisation the spend is meant to build. #### A reader running an independent studio A pushback on the meter itself. Unmetered subscriptions hid the miss-to-hit ratio, this reader argued, and now that token billing has exposed it the whole arrangement is starting to feel less like a service and more like a "consumer fraud". The "can make mistakes" disclaimers, in their view, are an insult given the prices. The argument has more force than the register: the case for paying premium prices for premium answers depends on the premium answers actually arriving, and on a credible way of knowing when they have not. The essay's prescription, a floor plus a budget plus the occasional tough conversation, presumes the user can tell good output from bad. For users who cannot, the meter is a cost without a benefit, and the cap might be the kinder policy after all. ### Community voice What readers were posting on LinkedIn this week. The thread running through them sits close to the essay: the human is the part that doesn't come in the box. Four voices, on value over scale, composure as a skill, the relational layer of AI, and the human capabilities AI makes scarce. #### John Middlemiss, founder, Empact **[John Middlemiss](https://www.linkedin.com/in/jmempact)**, founder of Empact, a consultancy on performance, AI and customer trust, went to London Tech Week and came back unconvinced by the headline race. Everyone was talking about scaling AI faster. His reflection: *"scaling AI is not the same as creating value."* The advantage, he reckons, comes from how trusted, emotionally intelligent and valuable the AI becomes when it acts for real people, not how quickly you roll it out. That's a useful thing to hold next to the gold-rush framing. The model is becoming the cheap part. What you point it at, and whether anyone trusts the result, is the part that doesn't come in the box. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7472551187599413248) #### Michael Nevski, Director of Global Insights, Visa **[Michael Nevski](https://www.linkedin.com/in/michaelnevski)**, Director of Global Insights at Visa, the payments company, pulled a line out of an old podcast conversation that's been sitting with him: *"Clients do not want the drill. They want the hole in the wall."* His point is about what survives when generative AI and the market shift weekly. Composure, judgement, the curiosity to work out what someone actually needs. He calls composure a skill, not a trait. Here's the bit worth carrying: the capabilities he's describing are exactly the ones a model can't read off you. They sit in how you decide, and you only get the benefit if you've worked out what they are. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7472695605346074625) #### Glenn Parry, professor, University of Surrey **[Glenn Parry](https://www.linkedin.com/in/glenn-parry-a44356)**, professor at the University of Surrey, a public research university, was at the CADE 2026 conference for a talk and AI film on **the relational layer** of AI governance. The work looks at what people *actually do* in an AI encounter, not what the policy says they should. That's the layer most governance frameworks skip straight past. They write the rules for the model and ignore the human on the other side of it. If the interesting question is what the person brings to the exchange, then the person is the bit you have to study, not the bit you can assume. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7472669033394413569) #### Fiona Eastwood, Global CEO, Merlin Entertainments **[Fiona Eastwood](https://www.linkedin.com/in/fiona-eastwood)**, Global CEO of Merlin Entertainments, the company behind Legoland and Madame Tussauds, wrote in the Mail on Sunday about getting young people into first jobs. The AI line is the one that stuck with me: as AI reshapes parts of the labour market, *"these human capabilities, communication, judgement and emotional intelligence, are only becoming more valuable."* She means it about school leavers. It reads just as true for everyone above them. The skills AI makes scarce are the ones that were never written down in a job description, which is exactly why they're hard to replace and easy to undervalue. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7471853909096108032) [Read the email ↑](#email-2026-06-20) [Image: LinkedIn carousel: Average by default] See all 8 slides [Image: Slide 1: . Tell the model who you are. Or it makes you average] [Image: Slide 2: ] [Image: Slide 3: . Ask several models the same question and they land on a different main argument 3% of the time. People do it 65% of the] [Image: Slide 4: . A model can sharpen one person's writing and flatten everyone's at the same time. Better alone, narrower together] [Image: Slide 5: . Without it the model optimises for plausible. With it, it can be useful to you, in this role, on this decision] [Image: Slide 6: ] [Image: Slide 7: . Fill the empty box, and average stops being the default] [Image: Slide 8: ] ## Edition #17: Ride the bike *13th June 2026* ### Letters Ethan's first week pulled a fuller postbag than usual, and a sharper one. Most readers wrote in agreement with the door-open hypothesis, but with their own evidence attached. A run of leaders working in advisory and consulting picked up the apprenticeship question and pushed it harder than the email had room for. Two readers from the creative industries asked a different question: not whether to hire graduates, but whether senior people are paying enough attention to what graduates already see. A request before the letters. ### Community voice What readers were posting on LinkedIn this week. The thread running through them sits close to the essay: the tool is the cheap part, the judgement around it is the job. Six voices, from innovation to insight to strategy. #### Henry Coutinho-Mason, futurist and author, The Future Normal **[Henry Coutinho-Mason](https://www.linkedin.com/in/henry-coutinho-mason-3689572)**, the futurist behind The Future Normal, a book on near-future trends, thinks most people are reading the AI-apps glut graph wrong. Use AI end to end and you get *"generic, undifferentiated, slop-like products"*. His map: AI won't give you the *0-to-1 spark*, and it can't do the final *7-to-10 polish* where craft and resonance live, but it's made for the **messy middle**, fleshing out early-stage ideas past the *"good enough to share with colleagues"* barrier. Measure it on ideas surfaced, tested and killed, not products shipped. Most ideas die unshared; he's pointing at the tool that gets them into the room. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7469699355378225152) #### Mike White, marketing and growth advisor, Lively **[Mike White](https://www.linkedin.com/in/mike-white-lively)**, a marketing and growth advisor to mid-market CEOs at Lively, a marketing agency, has stopped explaining AI through frameworks. The smartest operators he talks to *"have stopped describing AI as a system. They're describing it as a worker."* The intelligence, he writes, *"doesn't arrive ready-made. It has to be built"*, by a specialist who sets it up and someone inside the business who owns it day to day. Take the worker framing seriously and the rest follows: you'd onboard it, budget for it and expect an account of what it did with the time. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7470216743416696832) #### Maani Safa, CEO, Poppins Agency **[Maani Safa](https://www.linkedin.com/in/maanisafa)**, CEO of Poppins Agency, an innovation-led creative agency, listed ten AI tools living at the edge of the conversation, most of them free, many beating the household names at specific jobs. The list is useful; the last line is the keeper: *"The advantage was never just the tool. It's knowing which one to reach for, and having the taste to use it well."* Taste doesn't show up on the bill, and it's the only line item that compounds. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7470102268940423168) #### Dom Wong, co-founder and CEO, Pogo **[Dom Wong](https://www.linkedin.com/in/domwong14)**, co-founder and CEO of Pogo, a consumer research platform built on a purchase-verified shopper network, launched this week with $32m raised and a blunt opener: *"The dirty secret about consumer research? It's overrun by fraud."* Pogo points AI-moderated video interviews at verified buyers of a specific product, thousands at a time, and returns findings in hours. Worth noticing what gets scarce when fieldwork stops being the constraint: knowing which question is worth asking. The interviews are about to be the cheap part. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7470121182218919936) #### Helena Kosinski, Vice President, Luminate **[Helena Kosinski](https://www.linkedin.com/in/helenakosinski)**, Vice President at Luminate, the entertainment data company, and chair of MusicTech UK's advisory board, came back from SXSW London with one thread pulled tight: culture's funding problem is *"a lack of language, not a lack of intent"*. An urban regeneration panel said planners and financiers have no shared framework for culture's value; the launch of the Sound Investments report her team wrote heard the same about music tech, a capacity issue in the knowledge of the industry. The money exists. The translation layer doesn't. Whoever writes that vocabulary ends up directing the capital. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7470427396719075328) #### Hugh Derrick, executive coach, eatbigfish **[Hugh Derrick](https://www.linkedin.com/in/hugh-derrick)**, an executive coach at eatbigfish, the strategy consultancy known for challenger-brand thinking, summed up a panel on how strategy survives the organisation in one image: the best strategists are **cat-herders**. His line: *"The real job of a strategist isn't to own and protect their personal articulation of the strategy but to help others build a strategy that they feel like they own."* Go short or go home; sell to the heart, persuade in the appendix. A strategy people own outruns a sharper one they don't. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7470890568596832256) [Read the email ↑](#email-2026-06-13) ## Edition #16: The open door *6th June 2026* ### Letters One reader, [Conor](https://www.linkedin.com/in/comccart), sent a line that stuck with me: "the organisations that win won't have better AI, they'll have better questions." The gap between using the tools and knowing what to ask of them, he wrote, is widening faster than most leaders realise, and that is where the value is pooling. What unsettles him is how many businesses are automating before they have worked out what is actually worth keeping. Last week I asked what you can see coming [over the next five years](archive.html#email-2026-05-30), and a lot of you wrote back with super thoughtful answers. I am holding them. They belong in the "what's next" piece I am working on for a future week with Rob Wild. So Conor's is the only letter this week. Keep the predictions coming, though! [See what the community is posting ↓](#community-2026-06-06) ### Community voice What readers were posting on LinkedIn this week. A single thread runs through them and through the essay: when the machine does more, what stays ours? Four voices working the same nerve. #### Phil Leslie, Chief Technology and Innovation Officer, Cornerstone Research **[Phil Leslie](https://www.linkedin.com/in/phil-leslie)**, Chief Technology and Innovation Officer at Cornerstone Research, the economic and litigation consulting firm, cut through the noise of the jobs debate. Almost every public claim about AI and work, he argues, is a cheap signal: the labs want their tools to look world-changing and benign, incumbents want to reassure their people, the doom side wants an audience, consultants want to sell change. None of it necessarily false, none of it a credible signal either. His own read lands squarely on this week's essay. AI is a powerhouse at producing analysis, but the more it produces, the more human judgement is needed to make sense of it. Demand for that judgement is soaring while the supply stays scarce. He's started writing it up under a name that fits: the judgment bottleneck. [Read on LinkedIn ->](https://www.linkedin.com/posts/phil-leslie_almost-every-public-claim-about-ai-and-jobs-share-7468720503067869185-AfMM/) #### Julia Kenyon, co-founder, yuzu+co **[Julia Kenyon](https://www.linkedin.com/in/julia-f-kenyon)**, co-founder of yuzu+co, a business and human-performance coaching consultancy, and a former BBC global brand leader, pulled out the most striking number in a new productivity study: eighty-nine per cent of executives report no impact of AI on labour productivity over the past three years. Yet employees whose managers actively champion AI are far more likely to see gains. Her read is that the conversation is misframed, treated as a technology challenge when the evidence increasingly points to a human one. The organisations getting the greatest return, she argues, are the ones investing as much in the managers and human systems that make change stick as in the technology itself. [Read on LinkedIn ->](https://www.linkedin.com/posts/julia-f-kenyon_one-of-the-most-striking-findings-here-is-ugcPost-7468582725050937345-o22n/) #### Kristin Luck, investment banker, Oberon Securities **[Kristin Luck](https://www.linkedin.com/in/kristinluck)**, an investment banker at Oberon Securities and a board director across the data, insights and analytics sector, published the firm's first-quarter report and called a threshold crossed: enterprises have stopped experimenting with AI and started operationalising it. That shift, she writes, is reshaping valuations and accelerating consolidation. Buyers are concentrating capital on platforms that pair AI-ready data infrastructure, governance and proprietary data with recurring revenue. The flip side is the squeeze: the generalist data shops, the ones without a defensible data asset or a clear AI capability, are facing multiple compression. The market is beginning to price the difference between owning the judgement layer and renting it. [Read on LinkedIn ->](https://www.linkedin.com/posts/kristinluck_q1-2026-report-data-insights-analytics-ugcPost-7468579927001513985-eTV_/) #### Kristi Zuhlke, generative-AI and analytics founder **[Kristi Zuhlke](https://www.linkedin.com/in/kristizuhlke)**, a generative-AI and analytics founder, surfaced the pricing question every knowledge-work vendor is circling. She'd spotted a market-research agency advertising cheaper options because AI streamlines the work, and noted that brands increasingly expect prices to fall for exactly that reason. The agencies are caught: cut prices to reflect the efficiency, or hold them and keep the margin? It's the commercial edge of the same shift the essay traces. When the machine does the grunt, the value of expert work gets repriced, and nobody has settled on the new number yet. [Read on LinkedIn ->](https://www.linkedin.com/posts/kristizuhlke_here-is-a-market-research-agency-saying-they-share-7468416446654541824-ji6Y/) [Read the email ↑](#email-2026-06-06) ## Edition #15: How We Got Here *30th May 2026* ### Letters Last week's edition on Elliott's maths homework, and what learning means now drew a small but unusually thoughtful postbag. The dominant thread was generational: readers who had recently been juniors themselves, parents of teenagers, and senior leaders watching graduates change. The disagreement that opened up was not whether the method works, but whether the new method costs something specific in the underlying mastery. Several readers brought parallels from their own kitchen tables. #### A junior at a strategy consultancy, two years out of college Drew the generational line sharply, having lived both sides of it. Joined the firm when language models were still alien terms, watched the curve bend, and now sees new graduates and a younger sibling treating AI as a permanent intern that gets the assignments done. The reframe was less about productivity than about the cohort effect: the conversation in their world has shifted in the last few months from "use AI more" to "learn how to use AI as a support, not a substitute." Closed on a vivid image of a car being pushed down a slope, then braked at the moment it began to roll on its own. The braking takes more effort than the pushing did. #### A senior voice at the same consultancy, writing at length Took the homework story as the right model and then layered three concerns. One: a growing trend of treating AI as the ultimate source of truth, with quality control lagging. Two: a personal pattern of using AI to build in unfamiliar areas, hitting a wall, and reverting to the older tools, search, blogs, code forums, video, to fill the gap. The hands-on detour still has value. Three: Malcolm Gladwell's ten-thousand-hours frame, applied to professional judgment. Senior practitioners have done those reps across multiple technology waves; newcomers may not get the chance to. The macro question they raised, which David did not address directly last week, was the one to sit with. Individuals report higher day-to-day satisfaction. The aggregate productivity numbers have not yet shown up. A productivity paradox, in their phrase, like the early internet era. Worth watching whether the lag closes or not. #### A parent of a teenager, writing from professional services Recognised the homework scene exactly, with a sharper edge than the original. Their child, working on chemistry one evening, asked the parent and asked AI simultaneously. Slightly mis-specified the AI query, so the AI disagreed with the parent's correct answer. The child strongly preferred the AI's answer, would not back down, and only changed their mind days later when the chemistry teacher confirmed the parent had been right. The move worth naming is the one about authority. The new pattern is not just "use AI to check the answer." It is "treat AI as the authoritative voice in the room," displacing other adults the child trusted previously. Practical questions sit on top of that. The respect question sits underneath it. #### A senior director at a global professional services firm Read the edition as a piece about how the reviewers, not just the doers, have to adapt. Their summary, in their own register: a focus on skills for the future has to come with a willingness to review young people's work in the way the young people did it, not the way the reviewer would have done it. The implied second move is harder. If the test of good work is no longer "could you have done this yourself, alone, in a room, in an hour," the assessor has to develop a new instinct for what good looks like in a method that wasn't theirs. Otherwise, the review collapses into either rubber-stamping or rejection. #### A reader with the week's tersest reply Stripped the argument to a sentence: the difference between AI doing it for you and you getting it is the difference most are not making. Stands on its own. The Elliott story turned on that exact distinction. The point of the homework was not the answer, which Elliott had within seconds. It was that he could defend the answer afterwards. ### Community voice What readers were posting on LinkedIn this week. The through-line: the gap between individuals and organisations on AI adoption, and what fills it. Six voices picking at the same nerve. #### David Johnson-Igra, founder, Scribes Consulting **[David Johnson-Igra](https://www.linkedin.com/in/david-johnson-igra)**, founder of Scribes Consulting, an AI advisory for communications leaders, opened with the sentence the essay is built on: *"A year ago, I didn't think AI would impact my work. Then, within a few weeks, I was out of work, and everything changed."* He built the advisory inside that gap. Knowledge graphs, configurations, agents, tool integrations. Owned by the client, not rented from a vendor. The individual-vs-organisation adoption arc as a person, not a chart. He saw what was coming about five months after he could have, and acted the day after that. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7462182251842203649) #### Rahim Hirji, keynote speaker, author of SuperSkills **[Rahim Hirji](https://www.linkedin.com/in/rahimhirji)**, keynote speaker and author of *SuperSkills: The Seven Human Skills for the Age of AI* (Kogan Page, out July), did the count on his own manuscript. The word *human* appears 194 times. The word *AI* appears 56. He's calling this cohort **Generation Human**. *"The conversation about AI has already turned. It's no longer about what the machines can do. It's about what we humans need to be."* That sits alongside David's Generation One and Generation Two framing in this week's essay as the third axis. Not the model getting better. The job changing shape around the model. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7462028618785988608) #### John Gleeson, founder, Success Venture Partners **[John Gleeson](https://www.linkedin.com/in/johngleeson10)**, founder of Success Venture Partners, an early-stage fund and customer success community, hosted his third CCO/FDE Summit at SaaStr this week. The one line that stuck: *"we've moved from the buzz of AI to the playbooks for this next era. We're in a new operating environment, and our playbooks from the SaaS era don't work. We're redefining the metrics, the hiring profile, and the mindset to succeed."* That is Generation Two in a sentence. The model crossed a reliability threshold. The work didn't just speed up. The operating environment changed underneath it, and the org chart hasn't caught up. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7460748061917704192) #### Phil Leslie, Chief Technology and Innovation Officer, Cornerstone Research **[Phil Leslie](https://www.linkedin.com/in/phil-leslie)**, Chief Technology and Innovation Officer at Cornerstone Research, read the WSJ piece on the first AI-native graduating class and pulled out the line worth keeping: *"Working alongside AI tools has made critical thinking even more important than AI literacy."* He pushed it further. *"Junior people definitely don't have a monopoly on AI skills, and I question how meaningful the phrase AI-native is."* And the harder question the WSJ piece didn't ask: *"If companies are hiring junior people based on their AI capabilities, how is this being assessed?"* The hiring question that becomes load-bearing the moment "AI literacy" appears on a job description and no-one has a rubric to score it. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7465035135097159683) #### Dylan Jones, Chief Communications Officer, Bold Square **[Dylan Jones](https://www.linkedin.com/in/dylanpauljones)**, Chief Communications Officer at Bold Square, on the WSJ's piece *"Corporate America Is Starting to Ration AI as Cost Skyrockets"*. He set up the gap with two C-suite memos. Earlier in the year: *"Hey, AI is going to enable a bright new future for this business, and every single person in the company is going to help us ensure we're the most AI-forward organization in our space."* This month: *"Hey, can you cut down on your use of AI tokens as it's getting too expensive. We'll just SAY we're AI-forward, don't worry."* The performative version of organisational adoption: say it, don't fund it. The essay's "organisations slept" line, twelve months on: awake enough to issue the press release, not yet awake enough to pay the bill. [Read on LinkedIn ->](https://www.linkedin.com/posts/dylanpauljones_corporate-america-is-starting-to-ration-ai-share-7466092568309637120-Mb1M/) #### Yvan Boudillet, Music x Tech ecosystem builder [Image: The four skills to navigate the music ecosystem in the age of AI: Curiosity (N), Critical Thinking (E), Humility (S), Creativity (W). Compass-rose framing from Yvan Boudillet's panel at the Centre national de la musique.] *Four durable skills for the AI age, curiosity, humility, critical thinking and creativity, and they transfer well beyond the music world that named them.* **[Yvan Boudillet](https://www.linkedin.com/in/yvanboudillet)**, Music x Tech ecosystem builder, came back from a Forum on Employment and Skills at France's Centre national de la musique with the answer he and three co-panellists (from Believe, Sony Music, Music Tomorrow) landed on. The question: what skills are essential to navigate the music ecosystem in the age of AI? The four they settled on: **Curiosity, Humility, Critical Thinking, Creativity.** On critical thinking specifically: *"we are now navigating a permanent flood of contradictory information, tools that promise everything, and decisions that carry real consequences. The ability to slow down, evaluate, and choose consciously has never mattered more."* The list reads transferable. Replace "music" with almost any sector and the four still hold. [Read on LinkedIn ->](https://www.linkedin.com/posts/yvanboudillet_curiosity-humility-criticalthinking-share-7466063858260922368-7zm6/) [Read the email ↑](#email-2026-05-30) ## Edition #14: Kids these days *23rd May 2026* ### Letters Last week's edition on "What boards accept" drew an unusually argumentative postbag. Thirteen readers wrote in, and the centre of gravity sat squarely on the frame itself: is subtraction the right move for a board, or is it a distraction from the harder job of raising ambition? Two readers from the professional services world pushed in opposite directions on that question. A senior leader at a global publisher took a different angle, jolted by a single number. And a quiet cluster wrote about the email itself, the audio edition, and the discipline of cutting back on what one's own AI setup is being asked to do. #### A managing director at a global strategy consultancy Pushed back on the central frame. Boards, they argued, shouldn't be dictating at the level of dashboards, training assessments, or six-month-old benchmarks at all. Their lane is capital allocation and risk. The posture should be relentless pressure on ambition, and the real question isn't whether a Copilot rollout or a cost-takeout programme is on the books, but whether the business model itself is being reshaped or merely optimised. On risk, they argued the calculus has flipped: moving too fast or locking into too few vendors is now its own exposure. The answer they proposed is optionality by design, continually re-underwriting strategy as the technology moves. The board's job, in their reading, is to raise the ceiling, not to subtract. #### A partner at a different professional services firm Accepted the frame but extended it with a specific zero-sum mechanic. In their role they face what they called a more or less infinite to-do list, so any time AI gives back tends to flow into more of the same: catch-ups further down the marketing list, internal stay-in-touches that usually don't happen, lower-priority work that survives without a formal prioritisation. The richer move in their reply was a parallel with post-pandemic working patterns. The vast majority of meetings have stayed online, which has saved senior practitioners large amounts of travel time. Where has that time gone? Into more client meetings. The empirical net effect, they observed, is that everyone in the sector can now stay in touch with more clients, every relationship has become more competitive, and industry revenue has not grown to absorb the new effort. A reminder that productivity gains at the individual level can vanish at the industry level. #### A senior leader at a global publisher Reacted to a single number from the Three Things. The claim that half of organisations have redesigned end-to-end workflows around AI gave them a jolt and made them want to move faster than their own organisation currently is. They asked a sharp methodological question in passing: how, exactly, are firms defining "redesigned"? A useful caveat to hold against the headline figure, and one to remember when quoting it. They also lifted the skill-building idea from Try This into their own organisation's guidance, which is the kind of practical pick-up the section is built for. #### A senior leader in a consumer-facing business Wrote about a small but practical move that landed last week. The Try This suggestion to compress one's accumulated lessons into a personal review skill prompted them to consolidate their notes and files, with the result that their tool of choice no longer has to re-derive their preferences each time. The reflection that followed mattered more than the action: they suspected they had been overwhelming their main configuration file with too much, and were planning a return to hierarchies and efficiencies. A useful reminder that the discipline of subtraction the essay applied to boards also applies, in miniature, to one's own setup. #### A marketing lead at a global entertainment business Wrote in to say the audio edition was a welcome addition. Brief, but worth surfacing because it confirms a pattern visible in the listening data: a meaningful share of the readership prefers the spoken-word version on the school run or the commute, and the audio is no longer an afterthought. I'll be investing more time in making that sound even better in the coming weeks. #### A founder of a creative innovation agency Wrote about the personal-stories-plus-AI-leaders combination, and singled out the broader letters section as something valueable they don't see elsewhere. Headline coverage in the field tends to feature wins and tidy case studies. What's missing, in their reading, is the muddling-through: how people are actually struggling to integrate these tools into existing work. A vote in favour of keeping the failure modes and the half-finished experiments in view, not just the lessons that have already crystallised. Keep them coming! #### A senior music industry executive Forwarded a New York Times opinion piece on the recent wave of commencement speeches that took aim at AI, with the suggestion that the "view from the other side" was worth a look. The contribution wasn't a counter-argument so much as a reminder: the AI-positive frame the newsletter runs on is not the dominant register in the wider culture this month. Many universities, graduation stages, and the broader op-ed world are pulling the other way. Readers planning to make the AI case inside a sceptical institution should know what they're walking into. #### My Steadman co-founder, Tim Ryan Wrote with two references that landed exactly on the essay's spine. Dieter Rams' "less, but better," and Saint-Exupéry's line that perfection is attained not when there is nothing more to add, but when there is nothing more to take away. Both pulled the subtraction frame into a wider tradition of design discipline, and made the case that the right kind of less is not impoverished but more rigorous. Tim also observed that boards have always faced a temptation to add (committees, dashboards, frameworks) and that the discipline of removing is often the harder and more strategic move. Surfacing this as house voice rather than as anonymised reader, because the references travel and tighten the essay's frame. Beyond these, one reader sent a useful link to a podcast interview with a senior financier expressing a more cautious view on the pace of progress. As a body of correspondence, the week landed firmly on the central question: is the right board posture to refuse, or to demand more? ### Community voice What readers were posting on LinkedIn this week. The through-line: where the work actually sits and who's doing it, now that AI does more of the drafting. Six voices picking at the same nerve. #### Phil Leslie, Chief Technology and Innovation Officer, Cornerstone Research **[Phil Leslie](https://www.linkedin.com/in/phil-leslie)**, Chief Technology and Innovation Officer at [Cornerstone Research](https://www.cornerstone.com), the litigation-focused economic and financial consulting firm, posted a string of pieces on AI's place in expert work this week. The single sentence that lands hardest: *"I don't want an AI as a doctor or lawyer, but I definitely do want a doctor or lawyer that is using AI."* He extended the same logic to courtroom expert testimony, arguing against AI-as-expert-witness under Federal Rule 702 because *"expert testimony rests on accountability, oath, perjury, professional discipline. A system that cannot be sanctioned cannot meaningfully be cross-examined."* The framing he's pushing is Expert-Using-AI as the right unit: every conclusion, every method, every input still owned by the named human. The professional bears the risk; the tool does the work. It rhymes with what Elliott decided about algebra. The method that survives is the one you can sign your name to. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7460833860927381504) #### Adam Peruta, Newhouse School, Syracuse University **[Adam Peruta](https://www.linkedin.com/in/peruta)**, an academic at the Newhouse School (also part of the Steadman team), ran a study with Carrie Riby and Ipsos: 20 brand ads, half human-made, half fully AI-generated, tested against predicted business outcomes. The AI work *"looked credible"*. The human-made ads tested 14% stronger on short-term sales and 17% stronger on long-term brand health. His read on why: AI performed best when the brief was functional and direct, and fell apart when it needed *"a creative leap"*, emotion, storytelling, a point of view. The pattern transfers. AI's good at the bits that look like fluency. The bits where someone has to decide what's worth saying still belong to a person. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7462150575988649984) #### Elizabeth Oates, VP Consumer Insights, Molson Coors **[Elizabeth Oates](https://www.linkedin.com/in/elizabethknoxoates)**, VP of Consumer Insights at Molson Coors, the brewer, posted a one-paragraph confession this week: *"I used algebra at work today. Algebra."* She was solving for an exchange rate on a Post-it. No Google, no AI, no calculator. *"An honest-to-goodness solve-for-x."* Her middle school self, she said, would be absolutely stunned. The thread back into this week's essay is the same question Elliott raised over the kitchen table. The methods we were taught and the methods we use have come apart. Sometimes the old one still earns its keep. Sometimes it sits in a drawer for thirty years and surprises you on a Tuesday. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7457925369157820416) #### Yogesh Chavda, founder, Yogi AI **[Yogesh Chavda](https://www.linkedin.com/in/yogeshchavda)**, founder of Yogi AI and an adjunct at the Moore School of Business, came back from a keynote in Puerto Rico with a number worth holding onto: *"58% of consumers now use AI to research purchases. Only 14% trust AI to complete the transaction."* He calls the gap the *"trust cliff"*. The point isn't the number. It's where the negotiation now sits. For thirty years, marketers have studied what makes a person choose one brand over another. The buyer's still a person, but it's a person plus an algorithmic filter, and the filter doesn't recommend the best brand. It recommends the one it understands. Brand work, in that frame, is partly the work of being legible to a machine that's about to do the choosing. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7462265702758920192) #### Conor McCarthy, independent AI adoption consultant **[Conor McCarthy](https://www.linkedin.com/in/comccart)**, an independent AI adoption consultant who's been showing up across this newsletter's reactions paragraph in recent editions, posted a four-piece run this week on what AI adoption actually looks like from the inside. The line worth pulling out: *"We've been teaching AI adoption wrong. Not the prompts. Not the models. The framing. Call AI a 'tool' and people treat it like a hammer. Useful when you need it. Back on the shelf when you're done. Infrastructure doesn't work that way. You don't pick up electricity. You build around it."* He gave the gap a name in a separate post: *"trust debt"*, what accumulates when people across an organisation use AI informally and don't tell anyone, then a client asks how AI was used on their account. Both pieces sit on the same observation: the adoption story most leadership teams are telling is the visible half, and the invisible half is doing most of the work. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7460601442819665920) #### Ben Churchill, software engineering leader, Steadman **[Ben Churchill](https://www.linkedin.com/in/ben-churchill-6357a7a)**, a software engineering leader on the Steadman team, ran a structured evaluation across roughly 55 production tickets at an enterprise client and wrote up what landed. *"With a senior engineer reviewing in the loop, agents delivered 2 to 10x throughput. With the human removed, autonomous success collapsed."* The agents confidently produced code that compiled, passed linters, and was wrong in ways that took longer to find than to have written by hand. His broader observation across the most advanced AI-native engineering teams: they're planning more, not less. *"AI agents execute exactly what you specify. Not what you meant. Not what you intended. What you specified. At speed, a vague ticket doesn't produce a rough draft, it produces an elaborate, confidently constructed wrong result that takes longer to fix than it would have taken to build correctly."* The skill that becomes scarce isn't writing the code. It's writing the spec. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7440077915939074049) [Read the email ↑](#email-2026-05-23) ## Edition #13: What boards accept *16th May 2026* ### Letters Last week's edition on choosing well drew an unusually thick thread of replies. Twenty-four readers wrote in; most opened a different angle on the theme. The disagreement of the week came from two readers who pushed back from opposite directions. Several readers reframed the saved hour entirely. A reader from a quality-engineering tradition pointed out that the discipline of choosing well has a long, unglamorous history. The selection below runs broader than the email; eight letters, each paraphrased. #### A partner at a global professional services firm, writing from the Gulf Pushed back on the framing. Choosing well, they argued, has always been the work; the impressive people in any career are the ones with clarity of thinking, able to discern what matters from what does not, hold the big picture and the micro mechanics together, and stick to a priority once set. Their worry about the current moment is that modern organisations fall into "operational hectic," producing volume because nothing forces a market test. Without a customer willing to pay, the value of what gets built stays blurry, probably indefinitely. They closed with where they put their own saved hour: family, thinking time, and some deliberate boredom, on the grounds that nothing creates innovation like it. Beautiful. #### A senior leader mid-pilot of Claude Code Picked up the personal half of the essay. Found themselves going to bed later each night because the new tool kept opening doors worth walking through. The reflection was honest: just because the work can be done does not mean it should be. Choosing well, they wrote, is the next frontier. The letter sits cleanly alongside the person above as the other side of the same coin: one is worried about firms drowning in shiny fluff, the other is watching themselves do exactly that on a personal level. #### A reader writing from a Lean Engineering background Took the essay back to a discipline that has been arguing the same point for forty years. The Toyota method asks for five steps: identify value, work out how to deliver it, remove waste, make it flow, then revisit. The first and last are the ones almost everyone skips. Benchmarking, interviewing customers and line workers, sitting in the work before changing it, then asking what to change next once the change has settled. The change itself is the fun part. The diligence around it is dull, which is why so few do it. Their read: build is the fun, asking why and for whom is the work nobody wants. The analogy lands as a reminder that the discipline David is naming is not new; it is just newly cheap to ignore. #### A reader running an independent consulting practice Reframed the saved-hour question by changing its denominator. The cost of building never was the bottleneck, they argued. The shape of work has shifted: a task that used to begin with a blank document and a few hours of carving now begins with ten previous documents and a quick brief crunched through a coding tool to land a draft that is seventy or eighty percent of the way there. The check-and-perfect work to bring that draft up to standard, and to scrub the tell-tale signs of a first AI pass, takes proportionally longer than starting from nothing. Their point: the sunk-cost calculus changes. A freelancer can abandon a half-good draft for a better task. Inside a larger organisation, first drafts stack up and the thinking time gets squeezed. #### A reader whose mini frontier team is rebuilding its annual plan Took the essay as a prompt to share where they have just landed. The frontier function will keep working centrally on harder integration questions and the compliance issues that surround them. Most of the team is moving back into the flow of the business, building tools team by team rather than in a central pod: how major events are planned, how leads are generated, how a business unit operates day to day. The intuition is that team-level capability beats individual tool-building. They wrote that quality is rising on the same input, but that they cannot yet see real choices being made with the productivity dividend. They closed with a worry that resonated with the essay: how many organisations have great unused tools sitting next to harassed employees still being asked to absorb the previous wave of ideas. #### A fellow AI consultant Offered the metaphor that lodged for several readers afterwards: we have all just won the work lottery. The classic anecdote is that lottery winners often struggle: relationships fail, purposes collapse, the upheaval finds the cracks. The hope is that some find new purpose and thrive. Their reframe placed the essay's question inside that picture. Each of us is refactoring our lives post-AI, and the question of what to build is in fact a question about what each of us values. As we refactor, the values underneath get exposed, tested, hardened. The closing image, taken from the myth: the Midas touch. Be careful what you wish for. #### A reader whose work spans coaching and consulting Added a framework the essay had implied but not named. Most people know the right thing to do if only they take the time to think it through. Even when they explicitly know it, that is rarely enough to make them do it. Importance is necessary but not sufficient. The thing has to feel urgent: external scrutiny, fear of missing out, a precipitating threat, a near miss. Otherwise it sits on a to-do list and stays there. They suggested running the nineteen-project list through that lens: the people who thanked David for an insight but never acted on it have not decided against; they have not yet been forced to decide. The most frequent question from their own clients is "what are others doing," not "what should we be doing." That preference for cover lines up neatly with the importance-urgency frame. #### A reader writing about career time and family time Took the essay's six-people list and held it up against a longer horizon. They wrote about the trade between career time and family time and the way a working life can compound into something that, viewed from the end, looks like decades spent earning a few minutes. Their letter was the emotionally sharpest of the week. It read less as commentary on AI than as a quiet rebuke to anyone using a saved hour to keep doing the same thing. Several other readers wrote in on the same theme: a reader in enterprise architecture noted that high-value processes are still hard to find regardless of how cheap building has become; a reader in customer success flagged that "knowing what to build is always hard, regardless of the technique used to build it"; and a reader running a frontier project warned that durable, maintainable, composable software remains a cascade of corrections, even when the first version came out of a coding tool in an afternoon. The agreement was that the new bottleneck is judgement, not capacity. The disagreement was over whether organisations will manage to develop that judgement before they spend the productivity dividend on more of the same. ### Community voice What readers are posting on LinkedIn this week. The through-line: AI lands as practice, not toolkit. Five voices at five different altitudes of the same argument. #### Tim Ryan, Co-founder, Steadman **[Tim Ryan](https://www.linkedin.com/in/timothyryanuk)**, my Steadman co-founder, listened to Cal Newport, the Georgetown computer scientist behind *Deep Work*, on the Prof G Markets podcast this week and came back with a line he can't shake: *"AI is like bringing a forklift into the gym."* Derek Thompson, host of *Plain English*, paired it with his own coinage, **cognitive time under tension**. The bit worth carrying isn't that AI hurts thinking. It's that the CEO check, our Check, Edit, Own discipline, now has a second job. Quality control was always the obvious one. The workout is the second. Every push-back on a draft is the rep that keeps you sharp at the work the machine's doing for you. Skip the reps long enough and the cost shows up in two places: the next deck you send, and the person you've quietly stopped becoming. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7460330528962052096) #### Fiona Eastwood, Global CEO, Merlin Entertainments **[Fiona Eastwood](https://www.linkedin.com/in/fionaeastwood)**, Global CEO of Merlin Entertainments, the attractions operator behind Legoland and Chessington, came back from an MIT and Blackstone AI gathering with the takeaway that wasn't about the robotics. *"Just as important to success is senior leadership alignment, and treating AI as a business transformation agenda, not a collection of isolated experiments."* The organisations she watched landing real impact were doing three things: focusing on three or four highest-value domains with potential for 20%+ bottom-line impact, committing to end-to-end transformation rather than scattered use cases, and aligning strategy, operating model and leadership behind the priorities. The pattern Tim and I keep seeing in our own client work: the AI question isn't "what tools" any more, it's "what does the org need to look like to absorb them". [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7459969768167550976) #### Mackenzie Nordal, Co-founder, Atheni **[Mackenzie Nordal](https://www.linkedin.com/in/mackenzienordal)**, co-founder of Atheni, asked her three kids this week how they actually use Claude. The fifteen-year-old transcribes her tutoring sessions and cross-maps them against the GCSE learning objectives in her Claude project. The twelve-year-old's built an audiobook that teaches him the whole GCSE science curriculum in the style of a Lottie Brooks novel. The ten-year-old's Claude has set her a daily 20-minute circuit: non-verbal reasoning while planking, a 400m sprint, a maths drill, a riddle. Mackenzie's reframe lands harder than any of the ban-screens arguments doing the rounds this week: *"The problem was never the technology... it's always been the absence of anyone meaningfully teaching anyone how to configure it and make it work for them."* The schools debating whether to allow AI are arguing the wrong question. The lesson is the configuration. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7452253563537231872) #### Marie Robelin, Global Insights & Innovation, Unilever **[Marie Robelin](https://www.linkedin.com/in/marie-robelin-31411442)**, who leads global insights and innovation at Unilever, came back from two days of strategic foresight with the Copenhagen Institute for Futures Studies and brought back a line that catches what most AI business cases miss: *"AI can scan faster, surface weak signals, and connect dots at scale. But humans must set direction, intent and judgement. Otherwise we just automate today's blind spots, faster."* She closes on Eric Hoffer: *"In times of change the learners will inherit the world, while the knowers will be beautifully equipped for a world that no longer exists."* The bit worth carrying isn't that AI changes foresight. It's that the speed advantage cuts both ways. Faster insight on the right question is a gift. Faster insight on the wrong one is a tax. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7459325568551997442) #### Nick Graham, CMO, Vertemis **[Nick Graham](https://www.linkedin.com/in/npgraham)**, CMO at Vertemis and a former Kraft, GoDaddy and McCain brand leader, pulled one line from his recent podcast with Vineet Mehra, CMO at Chime, the US neobank: *"Today is the worst AI we'll ever have."* Nick's point is for insights teams. If AI can automate reporting, analytics, survey design and synthesis, then running research isn't the value any more. The **golden nuggets**, the deep human truths that connect brands to real needs, are. It's the same shift hitting every research-heavy function. The instinct to defend the old deliverable is strong. The work that survives is the work AI can't yet do, and you can. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7460267673218637824) [Read the email ↑](#email-2026-05-16) ## Edition #12: Choosing is the work *9th May 2026* ### Letters A light week for letters. Two readers wrote in to say nice things; David thanks them and will pick up on the substance again next week. A reader at a major broadcaster wrote in to say the weekly email is one of the best things they get each week. And separately, the second reader this month told David they read it at a child's Saturday-morning sports activity. Are you reading from a sideline somewhere? Reply and tell us. ### Community voice What engaged readers are posting on LinkedIn this week. The through-line: the people-and-process side, not the tools. #### Sameer Modha, Measurement Innovation Lead at ITV **[Sameer Modha](https://www.linkedin.com/feed/update/urn:li:activity:7454783910485426177)**, an analytics and ad effectiveness leader, shared a satire about a manager who'd been caught micromanaging staff and decided to put them in sealed boxes. The pivot: *"So why are we shouting at our AI machines? They are trained on all of human language and words. Is that all emotion free? No. Of course not."* His diagnosis is that "Killer Prompts" thinking treats the model like a vending machine. Insert words, vend rewards. The actual move, he writes, is to remember you already have the toolkit you need: *"human empathy."* A useful corrective to the prompt-engineering arms race. [See the post on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7454783910485426177) #### Sara Lloyd, Group Communications Director and Global AI Lead, Pan Macmillan **[Sara Lloyd](https://www.linkedin.com/feed/update/urn:li:activity:7454513654865145856)**, Group Communications Director and Global AI Lead at Pan Macmillan, the publisher, came back from the Bologna Book Fair AI Summit with three observations worth quoting. The fear is fading and real curiosity has replaced anxiety, with Paul Kelly, the CEO of DK, naming as most valuable *"those who understand process and aren't afraid to question how things have always been done."* Mary McAveney of Abrams Books made what Sara called the sharpest observation of the day: that *"it's easy to fall into binary positions on AI, and that's completely the wrong approach."* Sara's reading: all-or-nothing thinking is the biggest trap, and thoughtful publishers are finding a middle way. And human authorship, Sara notes, is becoming a quality signal, not a given. The publishing reframe lands beyond publishing. [See the post on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7454513654865145856) #### Tim Ryan, Co-founder, Steadman **[Tim Ryan](https://www.linkedin.com/feed/update/urn:li:activity:7458179501202714624)**, co-founder at Steadman (with David), shared this week's Microsoft Work Trend Index, which surveyed 20,000 knowledge workers across ten countries and analysed trillions of anonymised Microsoft 365 signals. The headline he lifted: culture, managerial support, and the way teams are run drive roughly twice the AI impact of individual mindset, with culture alone running 2.5 times stronger than the strongest individual factor. Microsoft names the bit most leadership teams haven't yet: *"blocked agency."* Capable people, ready to use AI well, stuck inside organisations that aren't set up to capture the value. Tim reckons it's higher than Microsoft's reported 10%; David does too. [See the post on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7458179501202714624) #### Phil Leslie, Chief Technology and Innovation Officer, Cornerstone Research **[Phil Leslie](https://www.linkedin.com/feed/update/urn:li:activity:7457818538142117888)**, Chief Technology and Innovation Officer at Cornerstone Research, an economic and financial analysis firm advising on commercial litigation and regulatory proceedings, picked up a piece by Neil Sahota arguing AI may be the only way courts survive their existing administrative load, never mind anything more ambitious. The line worth carrying across to other regulated sectors: *"AI isn't a substitute for the bench. It's a collaborator that cleans up the noise so that human expertise can be applied quickly and consistently."* Phil's framing extends the point. The first practical use of AI in any expertise-led service isn't agentic substitution, the kind where AI replaces a person. It's clearing the operational backlog so the humans can apply judgement at the rate the demand actually needs. [See the post on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7457818538142117888) #### Mike White, Co-founder and CEO, Lively **[Mike White](https://www.linkedin.com/feed/update/urn:li:activity:7457444104319172608)**, co-founder and CEO of Lively, a brand-experience and live-marketing agency, posted a story about two failed AI-to-human handoffs in one week. His own and a client's. Both businesses had a polished AI moment, then dropped the handoff to a person and watched the goodwill evaporate. *"Your customer doesn't separate the AI moment from the human one. They feel it as a single experience, and they judge your entire business on it."* His own confession sits underneath: *"I've done exactly this. Jumped at the tool before sorting the process underneath it. Assumed the technology would smooth over the cracks."* The line that carries: *"The better AI gets, the harder those gaps are to hide."* [See the post on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7457444104319172608) #### Conor McCarthy, Independent Consulting Adviser **[Conor McCarthy](https://www.linkedin.com/feed/update/urn:li:activity:7455530500162510848)**, an independent consulting adviser, posted the most actionable tip of the week. Open NotebookLM, Google's research-and-summarising tool, and feed it your LinkedIn profile, your CV, and your personal website. Then click "Audio Overview." Ninety seconds later, two AI hosts are discussing your career like a radio show about you. He's been running the exercise with people building their personal brand. *"Most of them cringe a little. Some feel unexpectedly moved. Almost all of them hear something they'd never thought to say about themselves out loud."* The point isn't the audio. It's that handing the narration to something outside yourself surfaces what you've stopped noticing. Worth a try this weekend. [See the post on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7455530500162510848) [Read the email ↑](#email-2026-05-09) ## Edition #11: The bill and the harness *2nd May 2026* ### Letters What readers said about Edition 10: "Rise of the auditors" #### What resonated * **Recognition of the auditor role.** Readers replied that they had already started doing AI work in ways designed to help the auditor (themselves) check it, rather than to do the work itself. The essay named what they had been doing without a label. * **The economics of flat-rate pricing.** A partner at a professional services firm pushed back hard on the implicit pricing model. When the AI does better work as it is used more, flat-rate prices have to break. The full critique opened the door to this week's essay. * **The outsourcing trap.** Several readers picked up on the same pattern: providers offer low prices to displace the old capability, then raise prices once the alternative is gone. Familiar from outsourcing cycles in other industries. * **The apprenticeship worry.** Several readers, including one partner, named the single critical question for any firm whose business depends on people developing judgment over a decade: where does the apprenticeship come from when the AI does the entry-level work? * **Trust as the surviving premium.** A counter-voice from a different professional services group was that firms with strong audit cultures absorb the shock fine. As production commoditises, the trust premium rises rather than falls. #### Points readers raised #### A partner pushed back hard on flat-rate pricing Their full critique landed on Sunday morning and prompted this week's essay. The argument, in their words: *"Flat-rate prices ultimately seem inappropriate. In a situation where often better — AI-produced work is better, outcomes are more effective — there's a natural incentive to do more rather than reach an absolute standard. The natural progression of many outsourcing-type things is for providers to offer low prices to start with, get customers used to the service to the extent that they disband their traditional old-school capability, then put prices up so that there isn't really much saving any more."* They then asked the question this week's essay turns on: *"What is AI really trying to achieve, better or cheaper?"* They also flagged the indirect costs nobody is yet pricing: *"How do these costs relate to the actual underlying direct costs of providing the service (including consideration of this being a massive investment phase where profitability metrics are confused, there is a race for customer numbers) and then also the indirect costs of environmental impact from energy use, etc."* #### A senior strategist named the auditor role they had already started playing *"Love this. The requirement for, and presence of, the auditor role is a great insight. I'm even finding myself doing AI work in ways that are built more to help the auditor (me!) check it, rather than to do the work itself."* Last week's framing landed for them not as advice to take, but as a description of what they had already started doing without naming. #### Hallucinations in, hallucinations out A long-time reader from European professional services wrote about a former colleague turned PE investor, now less convinced AI will make humans superfluous. The colleague's argument, as relayed: *"AI is now being trained by their own output, hallucinations in, hallucinations out. We are at danger of building on sand that goes unchecked."* The right answer for now is: not yet, and not where the frontier labs are focusing. But the question matters more as agents start consuming agent output. The auditor role becomes load-bearing for exactly this reason. #### The email-overload inverse of the productivity story A chief information security officer captured the flip side of the AI-productivity narrative, in their words: *"When I email someone a simple task, I get back a laundry list of AI-generated questions that are relevant, but still represent more for me to process. Worse, instead of doing the work themselves, some people hand back AI-generated output that I could have produced myself, and might not solve my problem."* The punchline: *"I am drafting this email in my own words, but will ask Claude to grammar-check and improve it, which, of course, will make it longer and more polished, but also more information for you to process."* #### The counter-voice: firms with audit cultures absorb the shock A leader at a different professional services group wrote the sharpest counter to last week's frame. Their read: firms with auditors at every level already, like theirs, have *"familiarity with managing the issue, notably of partners spending a lot of time reviewing"*. The shape of the future for their firm: *"Our future role will remain one of being trusted. Just how we do that, and what we can charge for it, will change. The brand as a symbol of a trusted service and group of people will become ever more key."* They closed less optimistically: *"Chaotic times with no leadership from governments to manage what's happening. Somehow I doubt that will prove to have been a good idea."* ### Community voice What engaged readers are posting on LinkedIn this week. #### The Broadcom-VMware parallel: what surrendering the layer above the model looks like **[Aakash Gandhi](https://www.linkedin.com/in/aakashgandhi)**, Partner at L.E.K. Consulting, a strategy consultancy, posted the cautionary parallel that anchors this week's harness argument. *"Broadcom's dramatic changes to VMware's licensing and pricing model. Over the past 18 months, we have witnessed one of the most significant strategic pivots in enterprise technology, with price increases reported between 200% and 500%, and some cases approaching 10× previous levels."* David's reframe: this is what surrendering the layer above the model looks like in the abstract. Buyers committed to a vendor's stack, the vendor changed terms, and there was nowhere to go that didn't cost more. The pattern repeats wherever the buyer doesn't own the harness. [See the post on LinkedIn ->](https://www.linkedin.com/in/aakashgandhi/recent-activity/all/) #### The buy/build discipline that prevents the Broadcom outcome **[Chuck Reynolds](https://www.linkedin.com/in/cyreynolds)**, also a Partner at L.E.K., set out the discipline that prevents the Broadcom outcome. *"Buy point solutions when the capability is standardised and not a source of differentiation. Reinventing AP automation or HR workflows rarely creates a strategic advantage. Build proprietary capabilities only where the insight, data, or decision logic is truly unique to your business. That's where AI becomes an advantage. For mid-market companies especially, undifferentiated AI spend is dangerous. Every dollar building something generic is a dollar not invested in what actually separates you from competitors."* David's reframe: harness logic in plain language for anyone making a buy/build call this quarter. [See Chuck's posts on LinkedIn ->](https://www.linkedin.com/in/cyreynolds/recent-activity/all/) #### "Agency > Agents", and a Belgian cardiologist's Claude Code workflow **[Henry Coutinho-Mason](https://www.linkedin.com/in/henry-coutinho-mason-3689572)**, a futurist and keynote speaker, used his recent SXSW talk *"Multiplayer Futures: Co-Creating A Vision For SXSW 2030"* to push three slogans worth taking seriously. First: *"Fewer People, Better Jobs"*, a re-framing of AI-induced job transformation via Klarna, a Swedish payments company that rebuilt its customer service around AI. Second: *"Agency > Agents"*, why AI agents will make humans more, not less, important. Third: *"Crowd-Powered Creativity"*, illustrated with a Belgian cardiologist who had built their own Claude Code workflows for medical research. David's reframe: the cardiologist is the harness in microcosm. Same Claude Code, same subscription tier, but a doctor with a few weeks of investment in custom skills got a workflow nobody else in their hospital had. That's the bet this week's essay is asking readers to make. [See the post on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7444365782127595522) #### "Renting someone else's audience, and paying more for worse signal" **[Colin Lewis](https://www.linkedin.com/in/colinlewis)**, a behavioural economist who writes the Robotenomics newsletter on automation and AI, surfaced an eight-word epigraph from Sam Khoury that sat with David all week: *"The companies that don't own their own content pipeline will end up renting someone else's audience, and paying more for worse signal."* David's reframe: applies one level above the model too. The companies that don't own their AI harness will end up renting someone else's, and paying more for worse output. The compounding goes to whoever holds the layer above the model. [See Colin's post on LinkedIn ->](https://www.linkedin.com/in/colinlewis/recent-activity/all/) [Read the email ↑](#email-2026-05-02) ## Edition #10: Rise of the auditors *25th April 2026* ### Letters What readers said about Edition 9: "The proxy break" #### What resonated * **Writing as identity and belonging.** The essay unlocked deeply personal stories. One reader described growing up treating correct English as a way of fitting into British culture, only to find AI stirring up the same anxieties about exclusion. Several others shared their own complicated relationships with writing and correctness. * **The missing language for AI feedback.** Multiple readers described the same awkward situation: receiving clearly AI-generated work from someone they respect and not knowing how to say so. The vocabulary for constructive feedback on AI-assisted work doesn't exist yet. You can say the work is confusing, but saying "check your AI" feels different. * **Craft versus mass production.** The most quoted reframe. One reader mapped it to clothing: Temu at one end churning out mass-produced garments, Huntsman hand-cutting the finest suiting at the other. AI enables mass production of ideas. We probably need both ends of the spectrum, but we need to proceed with care. * **Time-spent as the new proxy.** If polish no longer signals effort, does telling someone how long something took? One reader asked: is duration the replacement indicator for "thinking happened here, even if AI was involved"? * **The thinking is in the reading.** Several exchanges converged on the same point: the cognitive work isn't in the prompting. It's in catching what the AI gets wrong, knowing it's wrong, and fixing it. If you accept the first output, you've handed the thinking over. #### Points readers raised #### Language as belonging, language as defence A reader shared one of the most striking responses the weekly email has received. Growing up, they treated "correct" English as a way to belong to British culture. The obsession turned into a form of self-defence: be crisper and more correct to bat people and insecurities away. AI has stirred it all up again. They've caught themselves search-and-replacing em dashes from their own writing so colleagues don't accuse them of using AI. "Something about this revolution is forcing us to confront our own prejudices," they wrote. "And forcing me to reconfront mine." #### The feedback gap for AI-sloppy work A reader received a proposal from a consultant they use and respect. Clearly AI-generated and sloppy. They questioned how to indicate both that the work was sloppy and that the consultant should use AI better. "The reaction to AI rests in the extremes," they observed. "It is either nothing (they couldn't tell) or a flat out 'this is slop.' It has yet to develop that important middle ground for constructive feedback." A problem many readers will recognise. #### Don't use AI as the sticking plaster for perfection A reader who described themselves as someone who loves writing and loves words pushed back on the idea of one "correct" way. They shared a story about painting with their children at the weekend: the children kept trying to copy their drawing. They had to encourage them to draw their own feelings, how the wind felt, what they remembered. AI would have made the picture look excellent but would have missed the beauty and messiness of how they all felt. "Remove the fear of getting it wrong," they wrote. "Don't use AI as the sticking plaster to ensure perfection." #### Where will thinking-quality create value? A reader at a professional services firm mapped out four scenarios for where depth of thinking still wins. Value investors who don't need to convince anyone: they hold the key to action by deploying capital on the back of their own analysis. Strategy consultants who need to convince a board: harder, because clients may fact-check recommendations with AI, re-entering the sophisticated noise. Transaction due diligence: AI creates the document, another AI probes it, and eventually the consultant sells the algorithm. And across all industries: management teams flooded with great-sounding but potentially hollow analysis, needing either extreme specialisation or a trusted human advisor to navigate. "If the quality of thinking remains so important," they concluded, "then we should focus a lot on teaching people how to think clearly rather than 'what's the standard.'" #### AI was reinforced for corp-speak A reader who works on AI-based creative tools made a precise technical point. Language models weren't just trained on corporate writing. They were reinforced for it. That's the mechanism. Doubt and ambiguity are optimisation penalties in the training process. The model was rewarded for sounding certain and smooth. Their advice: "Let your humanity show. Embrace the doubts, the ambiguities. Showcase evidence that works against your own premise. Make it bumpy on purpose." #### Writing is the process of not understanding A reader quoted a line that captures the essay's central tension: "Writing is the process by which you realise that you do not understand what you are talking about." If AI does the writing, where does the realising happen? They asked whether it's in structuring and iterating on the prompt, or in the back-and-forth editing using the CEO principle. A question the essay raised but deliberately left open. #### Links readers shared ### Community voice What engaged readers are posting on LinkedIn this week. #### "Service as software, not software as service" **[John Gleeson](https://www.linkedin.com/in/johngleeson1/)**, who runs a customer success community and investment fund, met Marc Benioff this week. Every sentence came back to outcome-based pricing: the unit of value shifting from access (seats, licences, subscriptions) to outcomes (revenue recovered, deals closed, problems solved). Delivered autonomously by agents, priced on results, sold by the product itself. "If you can get that virtuous cycle, that is a home run." When the person who built the go-to-market motion every B2B company runs on tells you it's over, it's probably worth paying attention. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7448133912628887552) #### "Ship decisions, not decks" **[Nick Graham](https://www.linkedin.com/in/npgraham)**, founder of Vertemis, a research and analytics consultancy, argues that insights teams need to stop defining themselves by what they produce and start defining themselves by the business outcomes they unlock. From function to capability. From reporting to activating. From insights as output to decisions as output. "An insight is only an ingredient. The real value is the idea, choice or action it enables." [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7448049098773143552) #### "Your job is mostly not to get in the way" **[Dylan Jones](https://www.linkedin.com/in/dylancjones)**, co-founder of Bold Square, a communications and marketing advisory, picks up on Zuckerberg building an AI agent to help him be CEO. But the more interesting detail is Meta's internal message board where employees share AI tools they've built. "That feeling comes from individuals seeing their friends try new things, maybe get recognised for it, and excited conversations over the water cooler. It builds on itself rather than coming out of Project Best Bot." [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7441665491087060992) #### "Non-deterministic systems need determined outcomes" **[John Gleeson](https://www.linkedin.com/in/johngleeson1/)** again, this time on why Customer Success only exists because something is broken. AI is collapsing the three gaps CS was built to fill: product complexity, customer capability, and value alignment. But as those gaps close, new ones open. AI systems are non-deterministic, and the work required to ensure a successful outcome has gone up, not down. "That's where CS goes. Not away. There." The auditor argument, applied to post-sales. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7441967873897517056) #### "How do you squeeze wide innovation through a narrow algorithm?" **[Nadim Sadek](https://www.linkedin.com/in/nadimsadek)**, founder and CEO of Shimmr, an AI creativity company, returned from the Bologna Book Fair with one question he can't shake: asked by the Director of the Polish Book Institute during a conversation about AI and emancipated expression. The colours, the covers, the people, the ideas, and one number so large it reframes everything about where publishing and AI now stand together. His [full dispatch from the fair](https://www.linkedin.com/pulse/bologna-book-affair-nadim-sadek-k0ine/) is worth reading. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7450877484037881856) [Read the email ↑](#email-2026-04-25) ## Edition #9: The proxy break *18th April 2026* ### Letters What readers said about Edition 8: "What a day can do" #### What resonated * **Team-level tools over individual training.** The strongest thread. Multiple readers engaged with the argument that building shared AI tools as a team is more effective than training individuals. The "thirteen skills in one day" detail and the contrast with individual training sessions landed hardest. * **The "walk the talk" challenge.** A reader at a professional services firm asked directly whether the firm itself has rebuilt any team processes with AI inside them. The essay's closing provocation ("If the answer is zero...") was quoted back. * **Practical demand for skills.** One reader didn't just respond to the ideas. They immediately asked for help building skills for their own use cases: PowerPoint templates, executive summaries, client preparation. The essay's thesis validated in real time. * **"Forget teaching people to use AI."** A reader reframed the argument provocatively: instead of building generalised AI training programmes, use precious time with domain experts to teach AI to do their work. The sharpest strategic challenge from the replies. **The leaderboard as an incentive model.** A reader in the entertainment industry picked up on the ### Community voice What readers who've [engaged with this email](https://steadman.ai/newsletters/david/leaderboard.html) have been posting on LinkedIn this week. The common thread: polished output versus genuine thinking. **[Brett Danaher](https://www.linkedin.com/in/bdanaher-entertainment-science)**, a professor of economics and analytics at Chapman University, can't unhear something in his students' pitches: *"X is broken. That's the problem. We're the solution."* McKinsey-deck cadence in every deck. He calls it **AI-ambic pentameter**. What's worth sitting with isn't that founders are writing better. It's that polish and ownership might be inverse. The more fluent the delivery, the less the founder's own voice comes through. Everyone sounds good. Nobody sounds like themselves. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7445567365238665216) **[Helen Field](https://www.linkedin.com/in/helenkfield)**, a transformation leader at L.E.K. Consulting, a strategy consulting firm, uses The Killers lyric as a prompt: *"Am I human, or am I dancer?"* Her list of what stays human (delegation, clarity, collaboration, responsibility) isn't surprising. Her punchline is: *"Delegate tasks, NOT responsibility."* And then she lands it: *"Write your own LinkedIn posts. AI does not need to do that for you."* The irony of reading that advice on a platform drowning in AI-generated content isn't lost. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7444740751877394432) **[Nick Graham](https://www.linkedin.com/in/npgraham)**, founder of Vertemis, a research and analytics consultancy, and former SVP of Global Insights at Mondelēz, summarised a conversation with Clorox's Oksana Sobol that cut to the quick: *"Spend less time in the middle. The biggest value sits upstream in problem shaping and downstream in activation."* Most insights teams are still shipping decks. The irony is that AI makes decks even easier to produce, which means the middle grows faster than either end. The organisations pulling ahead aren't making better decks. They're spending less time on decks entirely. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7448049098773143552) **[Pavi Gupta](https://www.linkedin.com/in/pavigupta)**, a market research leader writing the Infinity Growth Loop series, keeps sharpening a distinction that matters more each week: are you using research for support or illumination? He calls the first one **insights slop**. The drunk-and-lamppost metaphor. Lazy surveys fielded to prove a case never created value. AI just makes them cheaper and faster to field. What he's circling is the same proxy break from a different angle: the research looks more professional than ever, but the thinking behind it hasn't kept pace. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7444886949439533056) **[Liam Cole](https://www.linkedin.com/in/liam-cole-a6788257)**, director at Poppins, a digital creative agency, went through a cull this week. Newsletters. Apps. Subscriptions. His diagnosis: *"I've been drowning in noise."* The volume of polished, AI-enabled content was stealing his presence with the people in front of him. It's the consumer side of the proxy break: when everything looks good, nothing stands out. His answer wasn't a filter. It was a delete key. Less stuff. More people. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7442860890665664512) **[Henry Coutinho-Mason](https://www.linkedin.com/in/henry-coutinho-mason-3689572)**, trend researcher and author of The Future Normal, shared the full video of his SXSW keynote "Multiplayer Futures." He anchored on EO Wilson's line about paleolithic emotions, medieval institutions, and god-like technologies. Three themes stood out: fewer people doing better jobs, agency over agents, and crowd-powered creativity. The phrase to hold onto is **agency over agents**. The question isn't whether AI can do the work. It's whether you're still the one deciding what the work should be. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7444365782127595522) **[Dylan Jones](https://www.linkedin.com/in/dylanpauljones)**, chief communications officer and managing partner at Bold Square, a communications advisory firm, noticed something about Zuckerberg building himself an AI agent: if the CEO of Meta is only now building one, this technology is still being figured out by the people closest to it. But that's not the real story. The real story is Meta's internal message board where employees share what they've built. *"Your job as leadership is mostly not to get in the way."* Culture builds on itself when individuals see friends trying things. It doesn't come out of "Project Best Bot." [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7441665491087060992) [See what readers said ↓](#letters-2026-04-18) ## Edition #8: What a day can do *11th April 2026* ### Letters What readers said about Edition 7: "What Is Your Organisation Actually For?" #### What resonated * **The production system vs human system framing.** This was the line readers quoted back most often. Several said it gave language to something they'd observed but couldn't articulate. One senior leader at a broadcaster picked it out and said it raises the deeper question: why do we work at all? * **Stated vs revealed preferences applied to organisations.** The economic concept landed hard with people who see the gap between what leaders say and what they protect. Multiple replies extended it: one argued that empire-building is a revealed preference too, not just attachment to relationships. * **The gravity metaphor.** The idea that reversion after training isn't resistance but gravity. Readers working on AI rollouts said it reframed their frustration. One described their organisation's planned messaging as being about "people as the key to our business" and saw the weekly email as validating that instinct. * **The loneliness of solo AI productivity.** The trade-off between working alone with AI (productive but lonely) versus working with colleagues (engaged but slower) resonated with people who've experienced both. One reader who works independently said it captured what they've hated most about recent years. * **Capability vs capacity.** A reader's distinction that capability sits in people but capacity lives in the collective. Even agentic AI, which is inherently about systems acting in concert, demands that organisations think larger and different, not just leaner. #### Points readers raised #### Machine or living organism? A reader at a professional services firm introduced a thought experiment from the philosopher and former chess grandmaster Jonathan Rowson. The question: is your organisation a machine, or a living organism? If it's a machine, you repair, optimise, and polish it. If it's a living organism, you feed, nurture, and grow it. They argued the edition touched on a cognitive dissonance: business language emphasises the machine metaphor, but people's lived experience treats the organisation more like an organism. Their challenge: if we think of AI as augmenting an organism we want to nurture, how would that look different from optimising a machine? #### Revealed preferences aren't only about relationships A reader offered a more sceptical reading. Revealed preferences aren't only about valuing relationships, they argued. Some people are empire-building, using hierarchy to serve themselves rather than the organisation. They identified three other forces slowing AI adoption: short-term goals that aren't yet disrupted by AI (the "crocodile closest to the canoe"); the absence of a concrete, three-dimensional vision of what an AI-enabled future looks like; and a general numbness to speculative negative scenarios after years of clickbait catastrophising. Their summary: "Not like you do it today" isn't enough to provoke specific action. #### Theory of the firm, Lean, and Goodhart's Law A professor connected the edition to academic "theory of the firm" literature: the resource-based view, the knowledge-based view, the dynamic capability view. Where does AI fit? They suggested the real question for many leaders is whether they're running a business or filling their day. In a follow-up, they drew a parallel to Lean manufacturing: Toyota's five principles for removing waste from production processes might be close to what's needed for AI deployment, but not identical. They also invoked Goodhart's Law ("when a measure becomes a target it ceases to be a good measure") to describe what happens when money becomes the goal rather than a proxy for value. #### Capability vs capacity A reader in India shared a striking incident. A colleague couldn't deliver an innovative AI solution, not because individuals lacked capability, but because the organisation lacked a team with the capacity to execute it together. The distinction they drew: capability sits in people, capacity lives in the collective. Even deploying AI effectively requires organisations to think larger and different first, not just leaner. #### AI as a capacity-builder, not a headcount-cutter A leader at an entertainment company connected the edition directly to their business. Their teams are engaged in repetitive manual processes where growth is pushing additional volume through workflows that can't scale. AI's role, they said, isn't to replace people but to free them from internal admin so they can spend more time building client relationships. The instinct to use AI as a capacity-builder rather than a headcount-cutter: that was the thread they pulled on. #### The loneliness of solo AI productivity A reader who works independently shared the sharpest personal response. Working alone with AI is lonely and uninspired. Working with humans is passionate and engaged, if a bit slower. They don't think the answer is "choose humans every time," but they're fairly sure it isn't "optimise for speed" either. The trade-off is real and underrated. ### Community voice What readers who've [engaged with this email](https://steadman.ai/newsletters/david/leaderboard.html) have been posting on LinkedIn this week. The common thread: judgment. **[Brett Danaher](https://www.linkedin.com/in/bdanaher-entertainment-science)**, a professor of economics and analytics at Chapman University, can't unhear something in his students' pitches: *"X is broken. That's the problem. We're the solution."* McKinsey-deck cadence in every deck. He calls it **AI-ambic pentameter**. What's worth sitting with isn't that founders are writing better. It's that polish and ownership might be inverse. The more fluent the delivery, the less the founder's own voice comes through. Everyone sounds good. Nobody sounds like themselves. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7445567365238665216) **[Phil Leslie](https://www.linkedin.com/in/phil-leslie)**, Chief Technology and Innovation Officer at Cornerstone Research, a litigation consulting firm, argues that judgment isn't pattern recognition. In litigation and M&A disputes, it's knowing which patterns to trust when the adversary is actively trying to discredit your analysis. *"The bottleneck isn't intelligence. It's skin in the game."* AI can synthesise a thousand precedents. It can't stand behind that synthesis in a deposition. The distinction that matters isn't smart versus not smart. It's accountable versus not accountable. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7442193901404344321) **[Pavi Gupta](https://www.linkedin.com/in/pavigupta)**, a market research leader writing the Infinity Growth Loop series, coined a term I think will stick: **insights slop**. DIY research tools make it so easy to field a survey that people are using them to validate decisions they've already made. *Using research as a drunk uses a lamppost: for support, not illumination.* The dangerous part isn't bad methodology. It's that the organisation now has a data point, which feels like evidence, behind a question that was never honestly asked. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7444886949439533056) **[Nadim Sadek](https://www.linkedin.com/in/nadim-sadek-23443210)**, founder and CEO of Shimmr AI, a publishing AI company, has a phrase for what happens when people use language models without pushing back: **cognitive surrender**. If you don't engage, question, debate the output, you're outsourcing the thinking itself. What I keep turning over is the direction of the risk. Most people worry AI isn't good enough. Nadim's point is that the bigger danger is when it's good enough that you stop checking. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7444658303458017280) **[Henry Coutinho-Mason](https://www.linkedin.com/in/henry-coutinho-mason-3689572)**, an independent trend researcher and keynote speaker and author of "The Future Normal", built a website for 80 executive assistants over lunch during a hotel keynote. Forty-five minutes. He's never built a website before 2026 and has now launched eight or nine. The point isn't that AI makes building easy. It's that the person closest to a specific problem can now solve it without waiting for anyone's permission, budget, or roadmap. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7441859361234178048) **[Helen Field](https://www.linkedin.com/in/helenkfield)**, a transformation leader at L.E.K. Consulting, a strategy consulting firm, asks the question that's been following me all week: *"Am I human, or am I dancer?"* Her list of durable human skills (delegation, clarity, collaboration, responsibility) isn't surprising. Her punchline is: *"Delegate tasks, NOT responsibility."* And then she lands it: *"Write your own LinkedIn posts. AI does not need to do that for you."* The irony of reading that advice on a platform drowning in AI-generated content isn't lost. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7444740751877394432) **[Phil Leslie](https://www.linkedin.com/in/phil-leslie)** (again), on the junior talent pipeline: *"The fix isn't restricting AI access for junior people. It's redesigning their work so that using AI and developing judgment aren't in tension."* He frames judgment as critical infrastructure. Disrupt the pipeline that develops it and you don't just have a training problem. You have a supply-side constraint on the most valuable skill in the market. This connects directly to what I wrote a few weeks ago about whether organisations should still hire graduates. Phil's answer is yes, but the work has to change. [Read on LinkedIn ->](https://www.linkedin.com/feed/update/urn:li:activity:7445205863663054849) [See what readers said ↓](#letters-2026-04-11) ## Edition #7: What is your organisation actually for? *4th April 2026* ### Letters What readers said about Edition 6, "The system and the surrender." #### What resonated * **Cognitive surrender was personal.** Readers didn't just agree with the concept in the abstract. Several described catching themselves doing it: accepting AI output without challenge, noticing their own verification discipline slipping, realising they'd started to trust the confident tone. * **The "plz fix" example polarised.** The law firm partner who types two words and gets expert output back prompted reactions. Some saw it as the future of professional work. Others saw it as the sharpest illustration of the surrender risk. * **Dead time and boredom.** The opening about Elliott's basketball practice, and the joy of filling dead time with productive AI work, drew pushback. One reader argued that boredom breeds creativity. Dead time is when the brain reboots. * **The apprenticeship question dominated.** Multiple readers, especially those managing junior professionals, raised the same concern independently: if AI handles the routine tasks that juniors used to learn from, where does the next generation develop judgment? This was the single most common theme. * **Leaders stepping back, not forward.** The detail about three CEOs choosing to step down rather than lead through AI transformation landed hard. Readers questioned whether these were growth-mindset failures or rational self-selection. #### Points readers raised #### "Google Maps for the brain" A reader at a professional services firm offered the sharpest metaphor of the week. AI is becoming like satellite navigation: a few clicks, brain off, follow the directions, and suddenly you've driven into a muddy field when you meant to be at a client meeting. The deeper concern: as agents gain the ability to send output directly to clients, the gap between "generated" and "delivered" shrinks to almost nothing. #### "How do you tell off an AI?" A reader caught their AI making a confident arithmetic error: calculating an 11-year compound growth rate on ten years of data, then insisting it was correct when challenged. The question that followed: with a junior analyst, you give feedback and they improve next time. An AI starts fresh every time. The institutional memory that makes professional development work doesn't transfer. #### "I built a PROMPT COACH for the Civil Service" A reader in government, inspired by the 2,000-word prompt example, spent hours building a set of instructions that encodes good judgment about their role and institutional context. Next steps: a QA prompt tool, then a co-pilot assistant. The system-building the essay described, applied to public service. #### "Where do associates learn critical thinking now?" A reader in the Middle East raised the apprenticeship problem directly. Three concerns emerged: AI reduces the number of reps juniors get with core tasks, it challenges the on-the-job development of critical thinking, and there are limited frameworks for how junior staff should learn differently now. It's a question ## Edition #6: The system and the surrender *28th March 2026* ### Letters What readers said about the previous edition. #### What resonated * **The slope/intercept framework dominated**: ten of 22 replies engaged with it directly. Several readers applied the graph to themselves, placing themselves on one line or the other. The language of the framework was widely adopted in replies. * **Load-bearing friction**: the argument that "not all inefficiency is waste" prompted readers to connect it to civil service design, governance structures, and accountability processes. The planning spreadsheet example landed hard. * **PwC's services-to-platforms shift**: readers at professional services firms asked directly what this means for their own organisations. The shift from billable hours to subscriptions provoked the most operational anxiety. * **The centaur chess inversion**: the finding that adding a human to a chess engine now makes it worse prompted readers to ask how long the current human-in-the-loop phase lasts in their own fields. ### Points readers raised #### "With very low mastery, they see a miracle. Those with deep expertise are more sceptical." An academic who has invested heavily in AI adoption accepted the slope/intercept framework but pushed back on its completeness. The lowest-intercept people show the fastest growth partly because they're uncritical: they "see a miracle and are the most excited." Some enthusiastic adopters weren't so great at their jobs in the first place and are hiding behind the technology. High-intercept people, meanwhile, understand failure modes and know how many things can go wrong. And yet, the same reader wrote two days later: "I still feel constantly behind and in danger of being passed." Someone who has invested heavily, agrees with the framework, and still feels vulnerable. #### "They need to support the changing of the workflows they don't see, but know, are critical." A reader at a media company challenged the implicit assumption that senior leaders should be using AI tools personally. The reframing: very senior leaders don't need AI in the same way more junior people do. The more senior you are, the more you are already handing off work to your "agents" (your team). The question for senior leaders isn't whether they log in more. It's whether they support the changing of the workflows that they don't see, but know, are critical to them getting the job done. Leadership, not tool adoption. #### "It's possible that the person in this is me." A reader invoked Sinclair: "It is difficult to get a man to understand something, when his salary depends on his not understanding it." Then applied it to themself: working hard for that not to be the case, but aware of the structural incentive to resist. Their harder question: if the people best placed to lead change are also the ones whose positions are most threatened by it, how does any organisation actually adapt? Their honest answer: probably by more people doing more things for longer than the automation narrative suggests. #### "It's not about time saving. That's the 10x game. It's about value add and surplus. That's the 1000x game." A reader challenged the graph directly, arguing it understates the amplification effect for already-capable people. The determining factor: what they called the "explorer mindset" (intellectual curiosity, creativity, constant learning), which "cannot be taught. It is self-discovery." The fear for their own organisation: "we run the risk of becoming the new average." #### An event as a test: planning strong, live operations untouched A reader in media described their biggest event of the year. Pre-event planning and post-event review were stronger than ever, with AI at the core. The week itself, however, was almost entirely unassisted by AI. Their own learning has been "episodic rather than continuous," with jumps in capability rather than a steady upward curve. Overall, the easy part: integrating AI into their own work. The hard part: building systems that stick, democratising knowledge, working within existing tools and infrastructure. #### "Many of our staff are non-native English speakers." A reader in government described an experiment: hiring someone with maximum AI flexibility to find pain points and build tools. The clearest win wasn't efficiency. It was helping colleagues write in English when many staff are non-native speakers. The stress-reduction benefits were as important as the productivity gains. A second observation raised the geopolitical dimension: Chinese AI models with access to Chinese social media offer capabilities Western-approved tools cannot match, but policy restricts integration into government systems. #### "Only variety can absorb variety." A professor of digital transformation extended the chess analogy. While AI alone now outperforms human-plus-AI in chess, "it's actually pointless for a computer to play another computer. The purpose of chess has stayed fundamentally with people." The deeper point drew on Ashby's Law: markets change, customer needs evolve, and AI models trained on historical patterns may miss novel situations. The learning growth curve matters because it builds the variety needed to respond to genuine novelty. #### "Adoption inflects when leadership links the tool to non-negotiable outcomes." A reader in learning and development ran deep research into past technology transformations (internet, email, SaaS). The key finding: adoption doesn't accelerate when leaders pitch "innovation." It accelerates when leadership links the tool to outcomes that cannot be negotiated away: safety, pay accuracy, service accountability, regulatory continuity. #### The fire-and-rehire question A reader in corporate finance shared a striking anecdote: a company told their firm this week that they had recently let go their entire technology and development workforce and asked them all to reapply for their jobs "with an AI lens, given the role had changed." The reader's framing: navigating a moving minefield, each user forging their own path. #### Three tensions that run through many replies A reader identified the three predicaments that kept surfacing: retaining senior roles with judgment while losing the apprenticeship pipeline that produces judgment. Foresight to expand versus extracting cost in the short term. Building capability by going deep versus experimenting with many tools due to fear of missing out. #### The curves should be exponential A reader suggested the slope/intercept lines in the graph should be exponential rather than linear: learning creates more ability to learn. The exponential version would be more accurate. And considerably more brutal. ### Links readers shared ## Edition #5: Reckoning and slope *21st March 2026* ### Letters ### What resonated * **The apprenticeship pipeline, again**: the question of what happens to junior roles when AI handles the volume work has now been the most-discussed theme across three editions running. This week it drew the sharpest language yet. * **The pace of change**: a partner at a consulting firm captured a feeling several readers seem to share: trying to "get on a breaking tsunami with a surfboard, and the surfboard keeps being reinvented." * **The ATMs-to-iPhone distinction**: the structural argument (automating within your paradigm vs replacing the paradigm entirely) prompted readers to apply it to their own organisations. * **The three-tool limit**: the BCG "AI brain fry" research resonated, particularly the finding that high performers were the first to be affected. ### Points readers raised #### "Porsches are stunningly quick and razor-sharp. A skilled driver can make one dance. A bad driver? They'll put it straight into a tree." A professor of digital transformation wrote an academic paper in two and a half minutes using an AI tool. Was it any good? No. Could it get published in a poor-quality journal with minimal tweaks? Yes. With nearly a hundred papers behind them, they know exactly what to add, what to remove, what's junk. "However a non-expert could do the same and wouldn't see the errors. An AI or non-expert reviewer wouldn't see the obvious error either and would accept it." The result is "lots of AI science slop" across academia, publishing and music. #### "Five years from now, the marketplace will offer nothing but blight." A reader at a professional services firm wrote: "I vacillate between being optimistic that AI will allow employees to contribute more vs. expecting that AI will bring mass layoffs and throw the world into desperation never before experienced." On the apprenticeship pipeline: "I simply cannot get past the shortsightedness of it." This from someone who describes being "fiercely AI curious" and learning in what little spare time there is, which makes the tension all the more real. #### "I'm trying to get on a breaking tsunami with a surfboard, and that the surfboard keeps being reinvented while I'm about to step on it." A partner at a consulting firm had been thinking about a comment I made in our last meeting that started "I wouldn't have said this two months ago, but..." The question: does this slow down at any point, or does ChatGPT just start to feel like yesterday's news forever? I don't have a comforting answer. I think the honest one is that the pace of change isn't going to decrease. #### Maintaining team size while expanding capability An IT director at a consumer brands company has been "advocating for internally as well: maintaining team size while leveraging AI to increase capability rather than running leaner." The argument: if growth is the goal, a team of several people using AI will be far more productive than cutting headcount and expecting one person to carry the load. A practical instinct too: "I've encouraged our team to avoid signing long multi-year contracts right now. The landscape is shifting so quickly that new competitors are appearing constantly." #### Substitute or complement? The ATM analogy goes deeper. A CTO had been working with a simple framework: "AI is a substitute for low-judgement work and a complement for high-judge work." But the ATM article complicated it. The key passage quoted back: "it is paradigm replacement, not task automation, that actually displaces workers." A more nuanced distinction than substitute-versus-complement alone. #### The explore-exploit tension in tool choice A data strategist pushed back on the "two or three tools" advice. "There's an explore/exploit conundrum of humans too but overall I'd say there's too much 'getting comfy with what I know' esp in the context of things getting better all the time." The nuance: "Like you I have settled on CC [Claude Code] but then building tools on top of that. So tool here is an interesting thing to define." One platform with many custom tools on top is different from three unrelated platforms. #### "What training or frameworks exist to roll out AI with care?" The AI lead at a major media company asked the question the essay left open: "I'd be interested in any training or frameworks you're coming across to roll this out organisation-wide with a consistent approach." A single sentence that captures what I'm hearing from senior leaders everywhere right now. The honest answer is that the frameworks are being built in real time, mostly by the organisations brave enough to try. #### Hiring juniors only matters if you care about legacy A colleague argued that investing in the next generation depends on whether leaders care about the company's future beyond their own tenure: "I would imagine hiring and training juniors only matters if you care about people, legacy or the company's future into the next generation. If you don't and just want to earn/sell in your lifetime then I guess they don't care and I'd imagine most don't." And: "I don't want to be a luddite but it does seem like as a civilisation we're not going in the best direction." ### Links readers shared ## Edition #4: The power and the care *14th March 2026* ### Letters ### What resonated * **The apprenticeship pipeline paradox**: the argument that cutting juniors today erodes the senior talent pool of tomorrow was the single thread readers returned to most. Multiple replies engaged with it independently, suggesting it articulates a worry many leaders already carry but haven't named. * **Extraction versus expansion as a choice**: readers responded to the framing as a decision, not a trend. Several said it sharpened conversations they were already having about whether AI headcount savings should be reinvested or banked. * **Skill reclassification in professional services**: the observation that AI has retroactively revealed which tasks were genuinely cognitive and which were merely time-consuming landed hard, particularly among people in consulting and law. * **The SaaS market pricing shift**: the 30% software stock decline and the "build it yourself" examples prompted readers to reconsider their own vendor relationships. ### Points readers raised A senior technology leader at a global professional services firm challenged the SaaS disruption premise. Development costs, they pointed out, are only about 20% of a typical software company's revenue. Sales, marketing and customer success absorb 60%. AI can rewrite code, but it cannot replicate distribution and switching costs. They drew a parallel to offshoring: "Huge appetite. Need for re-invention." The disruption is real but the mechanism is more nuanced than build-versus-buy. A partner at another professional services firm identified the tension between original thinking and process execution. Developing the foundational insight that makes a project valuable is still human work, they argued, but once that insight exists, AI can scale the execution. Their question: does this shift advantage or disadvantage people who trade on original judgment? "I suspect the answer is that it depends on whether I tool myself up appropriately." A founder building an AI-native company connected the apprenticeship argument to institutional culture. They started their career as a graduate trainee at a large bank and worry that the next generation won't get the benefit of those early years inside large institutions. They posed a sharp question: will we see geographic or cultural differences in AI adoption, where firms with cultures that already embrace apprenticeship end up moving faster? A professor setting up a Future of Work institute identified a specific parallel to the extraction-versus-expansion frame. Job applications have lost their friction: candidates send almost infinite applications using AI, and firms screen with AI. Both sides have lost out. The old friction forced applicants to think before choosing. AI removed it entirely rather than redirecting it somewhere useful. A chief people officer at a global firm read the edition twice: "first over the weekend and then again this morning." The apprenticeship pipeline and organisational change sections spoke directly to the tensions they navigate daily. Sometimes the most valuable signal is that a piece is worth re-reading. A newsletter author raised a fair editorial challenge: some sections sound more like AI than like me. "If it's your human insight, it jars a bit to read those bits in a voice that's obviously an AI." The same week, an executive I've worked with for years wrote to say that they loved how clearly they could hear my voice and personality in the writing. So one reader thinks there's too much AI and another thinks it sounds exactly like me. Either I've trained the model well or I've always written like a robot. I choose not to investigate further. ## Edition #3: Extraction or expansion *7th March 2026* ### Letters ### What resonated * **"The hundred small things" as a reframe**: the distinction between chasing dramatic AI wins and compounding small daily elevations. Several readers said it gave them language for something they had been struggling to articulate to leadership. * **The "extra hour" problem**: the observation that AI is deployed like an extra hour rather than an extra person struck a chord with people managing teams. Structures absorb the gain before anyone notices it. * **The junior roles question**: the Block layoffs and YC data generated the most emotional responses. Readers connected it to their own organisations' headcount conversations. * **The senryu competition as metaphor**: the retreat-versus-redesign framing resonated, though notably nobody offered examples of successful redesign. The absence may be the point. ### Points readers raised A senior leader at a global professional services firm identified a structural gap in the argument. The hundred small things need a container, not just encouragement. He proposed daily one-hour structured learning blocks rather than hoping people will explore on their own. His deeper point: senior leaders who do not use AI personally have no on-the-ground proof of benefit, so their teams see no credibility signal from above. An events and entertainment industry exec pushed the junior roles argument to its darkest conclusion. The UK already underinvests in training, preferring overseas hiring. If AI accelerates that trend, there is no junior pipeline to grow seniors from. "That is extremely bad for companies longer term in terms of skills shortages and salary premiums for skilled workers, and even worse for UK plc." A topic that I picked up in today's edition. A manufacturing executive used the framing to shape two specific conversations: accelerating superuser growth and celebrating a colleague's "let's map your process" approach to adoption stickiness. He is in the middle of major organisational expansion and sees the hundred small things as directly applicable to that work. A technology leader at a research firm noted a quiet loss that doesn't show up in any headcount data. His analysts used to walk to a colleague's desk when they got stuck on a coding problem. Now they ask AI. The problem gets solved faster. But the conversation that would have happened, the one where a junior person absorbs how a senior person thinks about problems, doesn't happen at all. AI is removing the apprenticeship mechanisms even where the apprentices still exist. My one proper unsubscribe turned out to be the most advanced person on the list. His world is so far ahead of ours that the weekly email isn't relevant to him. He works on cutting-edge AI implementation (sorry everyone, we're all just fast followers!). His AI communities are discussing running engineers 24/7 with 12-hour agent check-ins, deploying 30+ vibecoded projects from non-engineering teams, making openclaw work across 500+ person organisations, and polyphasic sleep schedules to optimise autonomous agent management. Anyone else even close to these conversations? For sure his world is a useful signal of where ours is heading. I'll stay close for you all! [Image: Polyphasic sleep patterns for AI agent management] *A glimpse of the frontier one departing reader lives on: sleep broken into shifts to keep autonomous agents supervised around the clock.* ### Links readers shared ## Edition #2: The hundred small things *28th February 2026* ### Letters ### What resonated * **The capability-adoption gap**: the tension between what AI can demonstrably do and what organisations will permit. This was the thread readers pulled on most. Thirty of 121 replies engaged with it directly, many in operational terms. * **"Back your misfits"**: the idea of finding and enabling the ten percent who don't need pushing. Several readers said they were already doing this and the framing validated their approach. * **Organisational inertia as structural, not cultural**: readers recognised the obstacle isn't attitude or skill but governance, process, and risk frameworks. One partner at a strategy consultancy pushed further: speeding up the same process is "premature optimisation." * **The personal-to-organisational transition**: the feeling of being ahead personally but constrained institutionally was widely shared. People described experimenting on weekends, then walking into Monday meetings where nothing has changed. * **"Package around problems, not platforms"**: a phrase several readers said entered their working vocabulary within days. ### Points readers raised A director at a media company connected the capability-adoption tension to their own industry. They are formalising a champions network, exactly the "back your misfits" approach, and said the framing helped crystallise what they were already building. An exec at a broadcaster shared the most striking story: they have been deliberately ignoring their organisation's AI policies to enable their team's experimentation. The weekly email validated an act of institutional defiance they were already committed to. An exec at a professional services firm offered a substantive counterpoint. Accelerating existing processes without redesigning them, they argued, is premature optimisation. Their real question: how does more information drive quality rather than just volume? A chief technology officer at a data firm placed themselves at stage seven of an eight-stage AI maturity framework they adapted from ---