# Steadman > Steadman helps organisations harness AI to work better, quicker, and happier. Human-first transformation from practitioners who deliver. ## Hero ### A steady hand in uncertain territory AI transformation isn't a technology problem. It's a human one. We help organisations build the skills, processes, and confidence to work better, quicker, and happier. --- ## Philosophy ### An electric bicycle for the mind AI is a powerful multiplier that still requires human steering. We believe in co-intelligence, humans and machines working together, not full automation. Question "we can automate that" with "but should we?" ### Three Pillars **Better** Higher quality work. Deeper insights. More thorough analysis. AI amplifies human judgment, it doesn't replace it. **Quicker** Hours become minutes. Days become hours. Not by cutting corners, but by removing the friction that slows great work down. **Happier** More time for work that matters. Less drudgery. Greater fulfilment. This isn't a nice-to-have. It's why transformation sticks. ### What's at stake Wasted investment in pilots that never scale. Governance failures that damage reputation. Competitors building capability while your programme stalls. The cost of getting AI transformation wrong is growing. The cost of doing nothing is higher still. ### Our Method **The 4 Ps** Preparation, Prompt, Process, Proficiency: the foundations of effective AI use that we embed in every engagement. **The CEO principle** Check, Edit, Own: all AI output must be verified, refined, and owned by the human. No exceptions. --- ## The Challenge ### Why AI programmes stall **The learning gap** Skills taught in workshops fade within weeks without sustained reinforcement. Knowledge transfer isn't capability building. **The pilot trap** Experiments that never scale. Foundational skills get skipped, so teams can't build on early wins. **The absorption gap** Time saved that never reaches the bottom line. Unless roles and teams change, the capacity AI creates is absorbed into busier days. This is the step almost nobody takes, and where most AI value has died so far. So we don't answer "how can AI help me" with AI. We ask which number you're trying to move, and work back from there. Where the value lives: [the AI Value Map](https://steadman.ai/newsletters/david/ai-value-map.html). --- ## Services ### Six ways we help ### Training & Transformation **Individual AI Coaching** One-to-one coaching for executives. Hands-on AI tool exploration applied directly to your actual work. Discovery call (30 mins) → Coaching session (90-120 mins) → Follow-up (2-4 weeks later). Most executives experience a breakthrough in the first session. The usual response: "Why did I wait so long?" *For: CEOs, CFOs, CMOs, Managing Directors, Partners, Founders, Board members* --- **Team Training & Transformation** Multi-week intensive programmes that build lasting capability, not just knowledge transfer. Week 1: 4 Ps methodology + hands-on exercises. Week 2: Effective prompts and team-specific "plays". Week 3: The CEO principle methodology and governance. Week 4: Complex workflows and impact measurement. Between-session homework, office hours, and detailed feedback throughout. Plus ongoing reinforcement, because knowledge without practice fades within weeks. *For: Transformation leaders, L&D teams, Department heads, Professional services firms* --- ### Strategic Advisory **AI Strategy Review** Rapid, focused engagements answering strategic AI questions. Board-ready deliverables. Phase 1 (Weeks 1-2): Leadership interviews, capability assessment, competitive landscape. Phase 2 (Weeks 3-4): Opportunity mapping, risk assessment, sector benchmarking. Phase 3 (Weeks 5-6): Strategic positioning document, prioritised roadmap, board-ready presentation. *For: Private equity due diligence, Board AI readiness assessments, CEOs preparing for investment/exit, Strategy teams* --- **AI Transformation Advisory** Ongoing advisory relationships keeping transformation programmes honest. Monthly 60-90 minute sessions plus written perspective on key decisions. Quarterly deep-dives on specific challenges. Ad-hoc access for urgent questions. We ask the hard questions others won't, drawing on cross-organisational pattern recognition. *For: AI transformation leaders, CDOs and CTOs, CEOs wanting ongoing counsel, Leadership teams navigating rapid change* --- ### Value Creation **AI Systems & Process Builds** Workflows, custom tools, and integrated systems built at the right sophistication level. - **Level 1: Workflows & Plays**: Structured processes with human in the driver's seat. - **Level 2: Custom Tools**: Purpose-built GPTs and assistants with context and guardrails. - **Level 3: Integrated & Agentic Systems**: AI embedded into business systems with human oversight at checkpoints. *For: Operations leaders scaling use cases, Product teams, Agencies productising expertise* --- **Embedded Transformation Support** Part-time, ongoing presence embedded with your transformation team. Typically 2-3 days per week for 6-12 months. We build playbooks alongside your team, coach individuals in the flow of work, challenge assumptions, surface problems early, and model what "good" looks like in practice. Transformation is organic. It requires presence, not just advice. *For: Programmes that have plateaued, Organisations building capability for the first time, Leaders wanting a trusted partner in the room* --- ## Track Record ### Evidence, not promises - **200+** Executive coaching sessions - **8** Books published - **18+** Months embedded in clients - **1000s** People trained ### Trusted by teams at BBC, an international strategy consulting firm, Expedia, Unilever 75+ years combined experience across EMI Music, BBC, L.E.K. Consulting, Direct Line Group, MasterClass, and Harrods. We understand how large organisations actually work. --- ## Professional credentials (shown on each team card) - **David Boyle**: 3+ years at forefront of enterprise AI transformation, 200+ executive coaching sessions, 50+ training sessions, co-author of 8 PROMPT books - **Tim Ryan**: 25+ years brand strategy and marketing leadership across cultural sector, transformation expert at Tomorrow London, EMI Music, Setanta Sports - **Sarah Clerkson**: over 20 years audience and customer insight across broadcast, streaming, telecoms, publishing, gaming and home entertainment; built Channel 4's 70-strong insight function; ex-Barb board - **Specialist Network**: Former executives from Unilever, BBC Studios, Channel 4, ByteDance, Kobalt, plus the former CEO of Urbanise ($100M IPO) --- ## Frequently Asked Questions **What is AI transformation coaching?** AI transformation coaching is one-to-one executive coaching that applies AI tools directly to your actual work. Our coaching sessions are 90-120 minutes where executives experience hands-on AI tool exploration with real business challenges. Most executives experience a breakthrough in the first session. **What is the 4 Ps methodology?** The 4 Ps stand for Preparation, Prompt, Process, and Proficiency. This is the foundation of effective AI use that we embed in every engagement. It ensures teams build lasting capability rather than just knowledge transfer. **What is the CEO principle?** The CEO principle (Check, Edit, Own) means all AI output must be verified, refined, and owned by the human. No exceptions. This ensures quality control and accountability in all AI-assisted work. **Who are the authors of the PROMPT books?** The PROMPT books were co-authored by Steadman's founders, particularly David Boyle. Published in December 2022, just 2.5 weeks after ChatGPT launched, these eight industry-specific books were among the first practical guides to working with AI language models. **How long does team training take?** Team training typically runs over 4 weeks with multi-week intensive programmes. Week 1 covers 4 Ps methodology and hands-on exercises. Week 2 focuses on effective prompts and team-specific plays. Week 3 covers the CEO principle methodology and governance. Week 4 addresses complex workflows and impact measurement. Between-session homework, office hours, and detailed feedback are included throughout. **What clients has Steadman worked with?** Steadman's team has worked with BBC, an international strategy consulting firm, Expedia, and Unilever. The team has delivered 200+ executive coaching sessions, been embedded for 18+ months in clients, and trained thousands of people in AI transformation. --- ## Team ### Practitioners who deliver Not theorists. Not hype-merchants. People with decades of experience who've done the work and built the capabilities themselves. Deliberately small and senior. No bench of juniors, no pyramid: the people who scope the work deliver it. ### Founders **Tim Ryan, Co-founder** 25+ years as a brand strategy expert and marketing leader. Executive marketing leadership roles in privately-backed growth and transformation businesses across the cultural sector, including Tomorrow London, EMI Music, and Setanta Sports. *Brings: Commercial strategy, go-to-market, training product development, partnerships* --- **David Boyle, Co-founder, Methods and Transformation** 25+ years leading insight and strategy inc. EMI Music, BBC, MasterClass, and Harrods. 3+ years at the forefront of enterprise AI transformation—200+ executive coaching sessions, 50+ training sessions and eight PROMPT books. *Brings: Executive coaching, embedded transformation, thought leadership* --- ### Directors **Sarah Clerkson, Insight & Strategy Director** Over 20 years turning audience and customer data into executive decisions, across broadcast, streaming, telecoms, publishing, gaming and home entertainment. Built and led Channel 4's 70-strong integrated insight function through successive reorganisations, and sat on the board of Barb. Previously Reach, Liberty Global, BT, Sky, PlayStation and Warner Bros. Home Entertainment UK. *Brings: Organisational design, audience insight, data strategy, evidence-based decision-making* --- ### Specialist Consultants **Richard Bowman** Consumer research, insight strategy - former Unilever and BBC **Niki Buitenrust Hettema** Global ML/AI infrastructure, strategy & execution – 15-year Googler **Ben Churchill** AI implementation, workflow tools, former CEO of Urbanise ($100M IPO) **Rufy Ghazi** Music technology and transformation - former ByteDance (TikTok) and AMRA (Kobalt) **Simon Jacobs** Analytics, performance measurement, former BBC Studios **Charlie Palmer** Brand strategy, creative industries - former BBC and Channel 4, managing editor of All 4 **Adam Peruta** Media technology, product development, Associate Professor, Newhouse School, Syracuse University --- ## Heritage ### From the authors of PROMPT In December 2022, just 2.5 weeks after ChatGPT launched, we published the first practical guides to working with AI. Eight industry-specific books helping thousands work better with language models. ### PROMPT Books (all available free) - PROMPT for Brands - PROMPT for Analytics - PROMPT for Startups - PROMPT for Musicians - PROMPT for Movies & TV - PROMPT for Fashion Retail - PROMPT for Podcasts - PROMPT for Real Estate --- ## How We Operate **Two-Person Rule** Every product area has at least two people genuinely interested. Creates healthy friction and ensures continuity. **No-Asshole Rule** We don't work with clients that make the team miserable. Selective by design. **10% Principle** 10% of our time goes to helping people who can't pay, including schools, charities, small businesses, and individuals. --- ## Contact ### Let's talk Whether you're exploring AI transformation, need a second opinion on your approach, or want to discuss what's possible, we're happy to have a conversation. **Email:** hello@steadman.ai No sales pitch. Just a genuine discussion about where you are and what might help. --- ## Company Information Steadman Partners Limited Registered office at Harwood House, 43 Harwood Road, London, SW6 4QP Company Number: 16979324 VAT Number: GB 512821911 © 2026 Steadman Partners Limited. All rights reserved. Created by Audience Strategies: https://audiencestrategies.com --- ## About Steadman Steadman is an AI strategy and transformation practice. We help organisations and their leaders use language models to work better, quicker and happier. Tim Ryan and David Boyle co-founded Steadman. Both spent 25 years or more in senior insight, strategy and marketing roles before this, at Tomorrow London, EMI Music, Setanta Sports, the BBC, MasterClass and Harrods. Sarah Clerkson joined as Insight & Strategy Director after over 20 years in audience and customer insight, including building Channel 4's 70-strong insight function. ### What we believe Most AI transformation failures are organisational, not technical. The models work. What stalls is everything around them: an unclear mandate from the board, tools bought but never set up well, no route from a promising pilot to standard practice, and nobody accountable for adoption. So we work on the operating model rather than the tooling. That means decision rights, governance, the capability to build, and the change programme that carries a new way of working into the business. ### What we do Six kinds of engagement, from a single session to a year: - Individual AI coaching: one-to-one executive sessions of 90 to 120 minutes. - Team training and transformation: four-week intensive programmes. - AI strategy review: six-week engagements producing board-ready recommendations. - AI transformation advisory: ongoing monthly advisory relationships. - AI systems and process builds: workflows, custom tools and agentic systems. - Embedded transformation support: a part-time presence for six to 12 months. Full descriptions of each are in [/llms-full.txt](/llms-full.txt). ### How we work We use language models openly and heavily in our own work, on enterprise platforms where client data is never exposed and never used to train models. A practice that advises on adoption should be visibly further along than the organisations it advises. Every output that leaves the team is checked by a person first. We call it the CEO Principle: Check, Edit, Own. A language model can draft. It cannot be accountable. ### Who we work with Consultancies, private equity firms and their portfolio companies, content and media businesses, and cultural institutions. Alongside the three of us, engagements draw on specialist consultants: Richard Bowman, Niki Buitenrust Hettema, Ben Churchill, Rufy Ghazi, Simon Jacobs, Charlie Palmer and Adam Peruta. ### Related Audience Strategies is our sister practice. Its AI work serves audience understanding and growth, where ours serves enterprise transformation. See [audiencestrategies.com](https://audiencestrategies.com). ### Company information Steadman Partners Limited. Registered office at Harwood House, 43 Harwood Road, London, SW6 4QP. Company Number 16979324. VAT Number GB 512821911. ### Get in touch Email hello@steadman.ai, or see the [contact page](/contact.html). ## Contact Steadman ### Email **hello@steadman.ai** reaches the practice and is the right address for anything commercial: an engagement enquiry, a speaking or workshop request, a question about how we work, or a correction to something on this site. **david@steadman.ai** is David Boyle directly, and is the address the Saturday weekly email is sent from. To subscribe, email him with the subject "Can you add me to your weekly email?" We answer enquiries ourselves rather than through an assistant or a form. There is no contact form on this site, deliberately. ### What to get in touch about - An AI strategy review for a board or executive committee. - Coaching for an individual leader, or training for a team. - A transformation programme that has stalled between pilot and standard practice. - A conference keynote or a hands-on session at a leadership offsite. - Press and media enquiries. If your question is about audience insight, research or demand measurement rather than enterprise transformation, our sister practice Audience Strategies is the better route: hello@audiencestrategies.com. ### Registered office Steadman Partners Limited Harwood House 43 Harwood Road London SW6 4QP United Kingdom Company Number 16979324. VAT Number GB 512821911. Post is received at the registered office but email is very much faster. ### Elsewhere - [David Boyle on LinkedIn](https://uk.linkedin.com/in/beglen) - [Tim Ryan on LinkedIn](https://www.linkedin.com/in/timothyryanuk) - [The Saturday weekly email](/newsletters/david/) - [Privacy policy](/privacy.html) and [cookie policy](/cookies.html) ## David's Saturday AI Thoughts *Page: /newsletters/david/* David sends a weekly email to people who are curious about AI. The archive page hosts five sections per edition: the email body, a LinkedIn carousel (cover image + full slides), "the bits that didn't fit" (interesting links and stories that couldn't quite make the cut), letters from readers (reactions to the previous edition), and Community voice (what subscribers who've engaged with the weekly email are posting on LinkedIn that week, with David's editorial take and links to each person's LinkedIn profile). Updated every Saturday. Deep-linkable sections: `#email-YYYY-MM-DD`, `#extras-YYYY-MM-DD`, `#letters-YYYY-MM-DD`, `#community-YYYY-MM-DD`, `#carousel-YYYY-MM-DD`. **Subscribe:** Email david@steadman.ai with subject "Can you add me to your weekly email?" ### The archive and its companions 29 editions from 2026-02-22 to 2026-09-05. The archive page (/newsletters/david/archive.html) opens as a wall of edition cards; each card opens one edition at reading width, with Newer and Older links. Editions carry date-keyed anchors, one per section family present in that edition: `#email-YYYY-MM-DD` (the email as sent), `#extras-` (the bits that did not fit), `#letters-` (reader replies), `#community-` (LinkedIn voices), `#try-this-` (prompts to try) and `#carousel-` (the LinkedIn slides). The same content, regrouped by type, is one file each: essays.md (29 essays), news.md (338 items), readers.md (49 letters and community sections), try-this.md (the prompts), and hub.md (this index). ### Artefacts One entry per artefact created for the weekly email: its summary, its opening, and the link to its markdown companion, which carries the page's prose in full; interactive results and data-driven panels live on the page itself. #### AI Value Map *Page: /newsletters/david/ai-value-map.html. Full text: /newsletters/david/ai-value-map.md. Published 2026-04-12.* Interactive tool that helps leaders map where AI value sits across four pillars of AI work and how much they're capturing, visualised as a Marimekko chart. Updated 2nd September 2026 from the original five-phase version. An interactive self-assessment tool from Steadman that helps leaders understand where AI value sits in their organisation and how much of it they're capturing. This is the updated version, published 2nd September 2026: it maps value across four pillars of AI work rather than the five phases of the [original April 2026 version](https://steadman.ai/newsletters/david/ai-value-map-v1.html), which is kept as a snapshot. ##### Intro Where's the value in AI for your organisation, and how much of it are you capturing? Which number are you trying to move? Name the measure, its starting point and who owns it, and judge the value below against it. Then this tool asks two questions. First: across four kinds of AI work, where do you think the biggest prize sits? Second: how much of that value are you actually realising today? It draws you a picture of your opportunity landscape. Takes about three minutes. ##### Step 1 of 2: Where's the value? Rate how much potential value you see in each of the four pillars. Zero means none; 100 means transformative. The tool will convert your ratings into relative shares. ###### The four pillars ###### Adoption Pillar 1: individual and team productivity People, and then whole teams, using AI competently in the work they already do: faster drafts, quicker research, shared prompts and habits that turn one person's trick into a team's standard. Raising the water line. The no-regret foundation, and the slowest to show up in the numbers. ###### Process re-engineering Pillar 2: real workflows rebuilt end to end Two or three real workflows rebuilt with AI at the centre, chosen by working back from the number in the P&L that has to move. The same headcount produces multiples of the volume, and the gains arrive at intervals you can plan around. *(The page continues; read /newsletters/david/ai-value-map.md for the whole of it.)* #### Organisational AI Diagnostic *Page: /newsletters/david/ai-maturity-diagnostic.html. Full text: /newsletters/david/ai-maturity-diagnostic.md. Published 2026-04-05.* Five assessment areas of organisational AI capability, grouped under the four kinds of AI work, with a 5x4 matrix, a 20-question diagnostic and result notes across Mindset, Strategy, Building, and Accountability. Five areas of organisational capability across four kinds of AI work, from the first person drafting an email with a chatbot to an organisation that works better, quicker, and happier because of it. Adoption is assessed separately for individuals and teams. The results show strengths, gaps and dependencies, not a place on a ladder. Most AI frameworks measure individuals. That is useful for hiring. But the question readers keep asking us is different: *where is my organisation strong, where is it weak, and what should it work on next?* This framework answers that. It uses four components — Mindset, Strategy, Building, Accountability — and applies them to five areas of capability, grouped under the four kinds of AI work: adoption (individuals, then teams), process re-engineering, roles and teams, and new products and revenue. It measures organisational maturity, not which tools people use — for that, see the [AI Usage Spectrum](/newsletters/david/ai-usage-spectrum.html). --- ##### Four things to look at, in every area An area is not a checkbox. In every one, ask the same four questions. Weakness in any one of them is what stops the work in that area paying off. ###### i. Mindset Do people treat AI as a tool, a threat, or a colleague? What does leadership model? Is experimentation rewarded or punished? ###### ii. Strategy Is there a deliberate view of where AI belongs in the business — or a hundred tactical decisions nobody has joined up? ###### iii. Building Can the organisation actually ship things with AI? Prompts, workflows, agents, products — what gets built, how fast, by whom? ###### iv. Accountability Who owns the output when AI did most of the work? How is quality checked? Where is the judgment that stops bad things shipping? --- ##### The framework, at a glance *(The page continues; read /newsletters/david/ai-maturity-diagnostic.md for the whole of it.)* #### The Five Questions *Page: /newsletters/david/five-questions.html. Full text: /newsletters/david/five-questions.md. Published 2026-07-18.* Five questions for turning AI capability into business results, each part of the work in its right court. *Five questions for turning AI capability into business results.* Published 18th July 2026. Most firms have done some parts of AI well, and most are struggling with the rest. Adoption continues while firms redesign processes, roles and products. These questions make the decisions and responsibilities explicit. This is the simplest model I can come up with, based on the firms I work with. It starts from the questions the work must answer. There are five. Answer them in order and the structure follows. ##### 1 · Strategy and vision **Which number must move, by how much, and who owns it? Is the strategy clear and well communicated?** The role AI plays in how we work (automation or augmentation), and the economic goal (efficiency or growth). Decided, written down, and communicated until nobody can miss it. [Map the opportunity against the business outcome](https://steadman.ai/newsletters/david/ai-value-map.html). **Whose court:** the board. ##### 2 · Tools: available, set up well **Does everyone have the current generation, set up well and properly funded, and is the next generation in pilot?** Available is the easy half. Set up well is the real test: the current generation in everyone's hands, configured properly, with the budget to use it fully. And the next generation in pilot, so that by the time it becomes the current generation the firm already knows how to use it. **Whose court:** the AI team for experiments and pilots; IT for everything in production, run as products. ##### 3 · Tools: used, widely and well **Are the tools used, and are there enough power users, amplified?** *(The page continues; read /newsletters/david/five-questions.md for the whole of it.)* #### The AI-Native Team *Page: /newsletters/david/ai-native-team.html. Full text: /newsletters/david/ai-native-team.md. Published 2026-04-25.* Interactive tool for modelling the Director, AI Builder, Auditor ratio. Diagnoses three gaps in how organisations staff AI work and provides a calculator for the team shape one AI Builder actually needs. Output has multiplied. Checking hasn't. Your best people are stuck doing the wrong jobs. ##### Three gaps ###### 1. Directors are drowning. AI has made output several times faster than a year ago. Reports, briefs, analyses, models: all arrive quicker. Directors are now reviewing more drafts than they can judge properly. The bottleneck isn't production. It's the queue of work waiting for a senior person to look at it. **Implication:** Take audit responsibility off them. Make sure work is checked before it reaches their desk. Check where direction and review become bottlenecks. The right capacity depends on demand, the work produced and the checking it needs. (See The Director, below.) ###### 2. The AI frontier is a full-time job. More than ever, staying at the frontier requires focus. The people doing the best AI work spend all day on it. They rebuild workflows every quarter and know which model fits which task. We should expect everyone to use AI. We can't expect most people to keep up with the frontier. Senior and client-facing staff are worth more on judgement and relationships. The frontier belongs to a specialist. **Implication:** Create a dedicated AI Builder role. Stop asking Directors to stay current on the tools and stop spreading frontier fluency thinly across the whole team. (See The AI Builder, below.) ###### 3. The checking is broken. Four failure modes, all common. Senior staff burning expensive hours on verification that sits below their pay grade. AI Builders pulled off the frontier to re-check their own output. Staff who look diligent but skim rather than check, passing errors through. Or worst: nobody checks at all. The work ships because the AI Builder is confident and the Director is busy. Assurance has been treated as a side-task. It's a craft. *(The page continues; read /newsletters/david/ai-native-team.md for the whole of it.)* #### Atoms of Enterprise AI Deployment *Page: /newsletters/david/atoms.html. Full text: /newsletters/david/atoms.md. Published 2026-05-05.* A reader's guide to the interactive Atoms diagram. Twenty-four atoms, one hundred and forty-five bonds, and a seven-link chain. A reader's guide to the interactive Atoms diagram. The diagram is a working map of how an enterprise AI rollout actually behaves. This companion explains what's in it and why each piece sits where it does. ##### What the map is Most enterprise AI rollouts treat themselves as a list of activities. Senior coaching done. Training delivered. Tools rolled out. Activities tick along while nothing changes. The map below treats deployment as a chain instead. Twenty-four atoms. One hundred and forty-five bonds. Seven links along a spine: a map of the dependencies between adoption and value capture, not a fixed order. Every link is only as strong as the audience it actually reaches. Business outcome, baseline, owner, selected workflows, capacity captured: that strip sits above the diagram. The map covers adoption and the changes needed to capture value in the existing business. New products and revenue are a separate kind of work, shown in the [Value Map](https://steadman.ai/newsletters/david/ai-value-map.html). Name the business outcome first, and [name the decision-makers and owners](https://steadman.ai/newsletters/david/five-questions.html); then inspect the workflow and role changes needed to achieve it, and the dependencies that might prevent them. It isn't a model that predicts outcomes. It's a structured way to argue about them. Every bond is author judgement, not a regression coefficient. Hover any atom to read its definition, its failure mode, and the diagnostic question that tells you whether it's broken. ##### How to read it *(The page continues; read /newsletters/david/atoms.md for the whole of it.)* #### AI: From What's True to What to Do *Page: /newsletters/david/ai-from-whats-true-to-what-to-do.html. Full text: /newsletters/david/ai-from-whats-true-to-what-to-do.md. Published 2026-04-05.* A structured argument for leaders that builds from three facts about what language models can do today, follows their implications to a decision only a leader can make, and maps the consequences that follow regardless of which route you choose. *Steadman. A Framework for Leaders. 5th April 2026.* This is a structured argument, not a briefing. It starts from three facts about what language models can do today, follows their implications to a decision only a leader can make, and maps the consequences that follow regardless of which path you choose. Each claim is tagged with its epistemic role — axiom, implication, your call, consequence, or recommendation — so you can see where the logic depends on evidence and where it depends on you. **Epistemic labels used throughout:** - **Axiom**: a statement accepted as true, serving as a starting point for reasoning. - **Implication**: a conclusion that follows logically from something already accepted. - **Your Call**: a decision, especially one that rests with a single person. - **Consequence**: a result or effect of an action or condition. - **Recommendation**: a suggestion or proposal as to the best course of action. --- ##### I. The Case — what's true today The opportunity builds from this. ###### [Axiom 01] Language models read, think, and write well. *axiom, noun. a statement accepted as true, serving as a starting point for reasoning.* Three core skills: **comprehension** (reading text and images), **synthesis** (reasoning, analysis, connecting ideas), and **writing** (including code, and therefore tool use). "Well" matters. The output is already useful, often good, sometimes excellent, today. ###### [Axiom 02] They loop to handle complex work. A model reads, thinks, writes. Then reads its own output, thinks further, writes again. Research, apps, analytics, computer control: all are read-think-write loops applied to different domains. *(The page continues; read /newsletters/david/ai-from-whats-true-to-what-to-do.md for the whole of it.)* #### Three Generations of AI *Page: /newsletters/david/three-generations.html. Full text: /newsletters/david/three-generations.md. Published 2026-06-03.* Chat, Agent, Employee. Three generations of AI that people can actually buy and use today, framed around what you're managing at each level. *Chat. Agent. Employee.* Three generations of AI that people can actually buy and use today. Each does more on your behalf than the last, and each asks more of you in return. The verb that runs through all three is *manage*. This is a partnership, not a handover. You are responsible for what the AI produces, every time, whether you're working alone or running an organisation. These generations describe delegation, not business value. Choose the setup for the task; use the [Value Map](https://steadman.ai/newsletters/david/ai-value-map.html) to decide which organisational changes matter. ##### Generation 1: Chat *You manage the execution of a task.* You ask, it answers. One prompt, one response, one decision about whether to use what came back. The conversation is the product. Everything that follows is a step up from this base. ###### 1A — Free / Constrained Capabilities and limits depend on the product, plan and configuration; check the setup people actually have. **Limited by design.** Builds foundational skill but sets a low ceiling on what people expect. **Examples:** Free ChatGPT, free Claude, basic Copilot, bundled tools. ###### 1B — Pro / Competent Good general-purpose chat applications. **Dramatically more capable** than the free tier: longer context, stronger models, useful features. **Examples:** Paid Claude, paid ChatGPT, paid Gemini. ##### Generation 2: Agent *You manage the execution of a complex process.* *(The page continues; read /newsletters/david/three-generations.md for the whole of it.)* #### The AI Usage Spectrum *Page: /newsletters/david/ai-usage-spectrum.html. Full text: /newsletters/david/ai-usage-spectrum.md. Published 2026-04-12.* A 3x3 framework mapping three kinds of AI delegation (Chat, Agent, Employee) against three levels of use, helping leaders see where their people are and where they need support. Rebuilt on the Three Generations frame on 5th September 2026. > The question for leadership isn't "are we using AI?" It's "which cell are most of our people in, and is that good enough?" Author: David Boyle, Steadman Published: April 2026. Rebuilt on the Chat, Agent, Employee frame 5th September 2026. ##### The framework Three kinds of delegation: Chat, Agent, Employee. Three levels of use within each: Poor, Good, Advanced. Most organisations have people scattered across all nine cells without knowing it. Choose the lightest setup that does the work well. Improve the person's skill within it, and widen its reach when the task and the controls justify it. The three kinds of delegation are the three generations in [Three Generations of AI](https://steadman.ai/newsletters/david/three-generations.html), each with two sub-tiers: Chat (1A free, 1B paid), Agent (2A sandboxed, 2B unleashed) and Employee (3A personal, 3B team). These generations describe delegation, not business value. ##### Chat (Generation 1) Examples: 1A free tiers: free ChatGPT, free Claude, basic Copilot. 1B paid: Claude, ChatGPT, Gemini. **Poor usage:** Generic briefs, unchecked answers. Treating it like a search engine, or owning a Ferrari and driving it in first gear. > Treating it like a search engine. "What is a SWOT analysis?" Getting a generic answer, pasting it into a slide, calling it done. Or, on a paid tool, using it exactly like a free one: never setting up custom instructions, never uploading a document, starting every chat from scratch with no context, accepting the first output without pushing back. Owning a Ferrari and driving it in first gear. On the free tiers (1A) the ceiling is lower and the habit is worse: the limits are so severe that the opinion people form is "AI is mediocre", and they stop pushing. *(The page continues; read /newsletters/david/ai-usage-spectrum.md for the whole of it.)* #### The CEO Gap *Page: /newsletters/david/the-ceo-gap.html. Full text: /newsletters/david/the-ceo-gap.md. Published 2026-06-20.* Every AI answer sits some distance from the correct one. This is a picture of that gap: how big it is, what it costs to close, and why the checking it needs keeps shrinking. *Check, Edit, Own. How the checking changes.* Published 20th June 2026. An AI has just handed you a draft. Do you trust it? Send it on, or check every line? You make that judgement dozens of times a day, and it decides how good your work is. The **CEO principle (Check, Edit, Own)** is how we make it today: by hand, on instinct. This page is the picture behind the instinct: how far an AI's answer sits from the right one, what it costs to close that gap, and why the checking keeps shrinking. The interesting question is whether it ever reaches zero, and for which tasks. ##### The four bands Every answer lands in one of four quality bands. - **Not good enough** — A wrong or off-target answer. Rarer now on a good tool: usually a vague brief, sometimes a real model mistake. Mostly avoidable. - **Good first draft** — In the right area, but it absolutely needs the full CEO. Cheap to get, expensive to trust as-is. - **Good enough to send** — Might carry the odd human-level mistake, the kind you'd make yourself. Light CEO, or just press send. - **Verified** — Known to be right, not just probably right: a calculation the machine re-runs and confirms, a quote matched word-for-word to its source. The specified property has passed an independent check; someone still owns the inputs, the scope of that check and the decision to use the result. ##### Chart 1: Closer and closer to the answer The AI's response climbs toward the correct answer as you put in more effort and spend (the horizontal axis is effort and cost: better prep, better prompt, better process). The space left above the line is **the CEO gap**: how much checking and editing the work still needs. Owning it never goes away. *(The page continues; read /newsletters/david/the-ceo-gap.md for the whole of it.)* #### The Four Freedoms *Page: /newsletters/david/four-freedoms.html. Full text: /newsletters/david/four-freedoms.md. Published 2026-08-04.* Four questions that tell you what any AI agent is actually free to do, using Cowork and Claude Code as the worked example. *Four questions that tell you what any AI agent is actually free to do* Everyone's first question about agent tools is "which one?". The better question is what any given tool is actually free to do. I coach people using Cowork, Claude's desktop agent, and people using Claude Code, its terminal sibling. They run the same engine; what differs is the containment. The difference comes down to four freedoms. Ask these four questions of any agent setup and you'll know exactly what you're working with, and what you're not. ##### 1. Reach — Where can it go? [Place this setup in the wider framework](https://steadman.ai/newsletters/david/three-generations.html). An agent's usefulness rises with how much of your world it can see at once. The real work of a good session is cross-referencing: the client file against the meeting notes against the thing you wrote in March. If the agent can only see one folder, you have to predict which connections it will need and copy everything in before you start. Most of the value lives in the connections you didn't predict. * **A contained setup (Cowork):** The folder or folders you granted, and nothing else. You prepare the room before the meeting. * **A setup with wider access (Claude Code):** The whole machine. Any project, any archive, any old note, connected in a single pass. ##### 2. Tools — What can it wield? Both can write and run code, which surprises people. The question is what the code runs against. Inside a sealed virtual machine the agent works on copies: safe, contained, disposable. With a real terminal it works on the world: your email drafts, your accounts, your website, your live systems. The distinction isn't capability. It's consequence. *(The page continues; read /newsletters/david/four-freedoms.md for the whole of it.)* #### Getting Executive Teams Hands-On with AI *Page: /newsletters/david/executive-sessions.html. Full text: /newsletters/david/executive-sessions.md. Published 2026-07-23.* A planning guide for AI sessions at executive retreats and leadership conferences, six goals, what works and what doesn't, and six formats, from three and a half years of running them. What works, what doesn't, and six ways to run it: a planning guide for retreats and leadership conferences, from three and a half years of running them. By David Boyle, Steadman (steadman.ai). 23rd July 2026. From the thinking behind David's Saturday AI Thoughts: https://steadman.ai/newsletters/david/ Our first AI session for a leadership audience ran on 2nd February 2023, nine weeks after ChatGPT launched: a hands-on workshop at a global consumer goods brand's international leadership conference, with executives generating campaign images for their own products using tools most of the room had never opened. This month the same company came back to start planning the next one. We have run these sessions continuously since: for an international strategy consulting firm, for boards, for private equity firms, for broadcasters, for founders. The requests increasingly sound the same: "our leadership team gathers once a year; how do we use that time on AI well?" This page is our answer. Use it to plan your own. And if we can help, we will run it with you. ##### I. Start with the goal, not the agenda The most useful question in planning one of these days is about purpose: what do we actually want to achieve? Most sessions are designed backwards, agenda first. The agenda is the last thing to decide. Which business decision must this day enable? Name the outcome it affects, the person who owns it and the next step the room must agree ([choose the decision this room must make](https://steadman.ai/newsletters/david/five-questions.html)). A leadership session can do six jobs. Pick two. A day that tries to do all six does none. *(The page continues; read /newsletters/david/executive-sessions.md for the whole of it.)* #### The Environmental Cost of AI *Page: /newsletters/david/ai-environmental-cost.html. Full text: /newsletters/david/ai-environmental-cost.md. Published 2026-07-25.* The full workings behind the essay "Burned and earned" — what a prompt, a task and a day of AI cost in energy, carbon and water, every number sourced and checkable. *The workings behind the essay, every number sourced.* This page accompanies the essay [Burned and earned](https://steadman.ai/newsletters/david/archive.html#email-2026-07-25) (25th July 2026). The essay makes the case: the cost of using AI is the footprint of the machine minus the footprint of whatever would otherwise have done the work. Here are the full workings behind every number in it, the model, the ranges and the sources, so you can follow the arithmetic or argue with it. ##### What a run costs ###### Generation 1: Chat, a prompt *A question, or "synthesise this document".* * Money: under a penny * Energy: 0.24 Wh (nine seconds of television) * Carbon and water: 0.03 g CO2e, 0.26 ml (five drops) * Human-energy floor: about 12 seconds The one measured anchor: [Google's published figure](https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference) for a median text prompt, covering chips, host machines, idle capacity and cooling. Google also reports that this figure fell 33-fold in a single year. ###### Generation 2: Agents, a task *Research a topic, build a model, make a deck.* * Money: about £1 * Energy: ~65 Wh (range 25–250) * Carbon and water: ~8 g CO2e, ~70 ml (a few sips) * Human-energy floor: about 50 minutes By energy, a task is a couple of hundred prompts. Not hundreds of questions: one long, growing conversation the model re-reads on every step. ###### Generation 3: All-day AI, a day *Triage the inbox, watch the channels, run a few tasks.* * Money: a few pounds * Energy: ~0.25 kWh (the kettle, boiled twice) * Carbon and water: ~30 g CO2e, ~270 ml (a mug) * Human-energy floor: about 3.3 hours *(The page continues; read /newsletters/david/ai-environmental-cost.md for the whole of it.)* #### Futures and Fears: The Full Record *Page: /newsletters/david/futures-and-fears-full-record.html. Full text: /newsletters/david/futures-and-fears-full-record.md. Published 2026-07-18.* What around twenty readers see coming over the next five years, the fears that arrived in one week's postbag, and what a four-person practitioner panel (Rob Wild of L.E.K. Consulting, Somnath Biswas of The AA, Scott Breitenother of Kilo, and David) said about AI that remembers, perceives and acts on its own. The full record behind the essay. Around twenty readers wrote down what they can see coming over the next five years. More wrote in this week with the fears that keep them up at night. And on Wednesday afternoon, a panel of practitioners spent an hour in public on what happens when AI remembers, perceives and acts on its own. The essay of 18th July 2026 distils all of it ([read the essay: Futures and fears](https://steadman.ai/newsletters/david/archive.html#email-2026-07-18)). This page keeps the detail. ##### I. What readers see coming The 30th May edition, "How We Got Here", closed by asking readers what they can already see coming. Around twenty replied, from consulting partners and bank executives to publishers, founders, investors and people early in their careers. Their answers cluster into eight themes. **The costs and the benefits land on different people.** The sharpest challenge in the postbag: the question that matters is who bears the cost of others acting, "as who drives the future of AI and who it most affects are not the same." Inside firms, the same worry in miniature: those already fluent will spend more and pull further ahead. **Judgement: what erodes, and where the next generation's comes from.** ([the longer history of this argument](https://steadman.ai/newsletters/david/weve-had-this-conversation-before.html)) The single biggest cluster. The risk sits in the boring middle of work, handed over because the machine's version is more thorough on first pass. "The line moves, but nobody moved it. We just never stopped to ask." And if junior work disappears, where does senior judgement come from? The proposed answer: deliberate, structured apprenticeship in judgement, not cheap execution. *(The page continues; read /newsletters/david/futures-and-fears-full-record.md for the whole of it.)* #### How the AI Story Was Told *Page: /newsletters/david/how-the-ai-story-was-told.html. Full text: /newsletters/david/how-the-ai-story-was-told.md. Published 2026-06-13.* Three and a half years of AI coverage across a ten-title media spectrum (MIT Technology Review; the Financial Times and The Economist; The Guardian and the New York Times; Harvard Business Review, the two MIT Sloan titles, McKinsey and BCG), read in full text where available and charted against the AI-newsletter wave and this email's own arguments, with a switch to flip every chart between four media tiers and the individual publications. A news-analysis retrospective created for David Boyle's Saturday AI Thoughts. It reads the whole record of AI coverage since ChatGPT at once: 20,637 articles across a media spectrum, arranged in four tiers — technical (MIT Technology Review); business news (the Financial Times and The Economist); mainstream news (The Guardian and the New York Times); and management and consulting (Harvard Business Review, MIT Sloan Management Review, MIT Sloan's Ideas Made to Matter, McKinsey and BCG) — plus the AI-newsletter wave and the recurring arguments of this email itself. A switch flips every chart between the four tiers and the individual publications. Most titles are matched on full article text; the New York Times (headline and abstract) and MIT Sloan Management Review (headline and teaser) are metadata-only and drawn dashed, and BCG is a partial capture. **Status: Raw AI output. Not yet CEO'd (Checked, Edited and Owned).** Page updated 8th August 2026; the article counts on the page sum to 20,637 across ten publications, with coverage and matching methods differing by title (some sources carry metadata rather than full text). The data cut-off is [unverified] beyond the page's own month axis. ##### I. The story that hasn't peaked Monthly AI article counts per title. A news story usually spikes and decays; this one is still climbing three and a half years on, with no peak in sight. An indexed view sets each title to 100 at ChatGPT's launch (30th November 2022) and compares growth on a log scale — a title that started near zero, like the New York Times, can grow many times over. The monthly journals run at lower volume than the dailies by design, so indexing is the fairer comparison of who accelerated. ##### II. First contact *(The page continues; read /newsletters/david/how-the-ai-story-was-told.md for the whole of it.)* #### We've Had This Conversation Before *Page: /newsletters/david/weve-had-this-conversation-before.html. Full text: /newsletters/david/weve-had-this-conversation-before.md. Published 2026-03-15.* Every generation warns that new technology is making us stupid; the concerns are not wrong, but they are not new either. You've probably heard someone say that AI is making us stupid. That we're outsourcing our thinking. That something fundamental is being lost. They might be right. But here's my view: **what we learn changes, not whether we learn.** The tasks are different now. The requirement to develop judgement is not. The reps still happen. They're just different reps. Critical evaluation rather than synthesis from scratch. It's also worth knowing that we've had this exact conversation before. Many times. --- ##### Writing · c. 370 BC > "This invention will produce forgetfulness in the minds of those who learn to use it, because they will not practice their memory… You offer your pupils the appearance of wisdom, not true wisdom." Socrates, in Plato's *Phaedrus* ##### The Printing Press · 1481 > "Abundance of books makes men less studious; it destroys memory and enfeebles the mind by relieving it of too much work." Hieronimo Squarciafico, Venetian editor ##### Newspapers · 18th century The French statesman Malesherbes railed against the fashion for getting news from the printed page, arguing that it socially isolated readers and detracted from the spiritually uplifting group practice of getting news from the pulpit. ##### Novels · 1799 > "They impair the mind's general powers of resistance, which lays the mind open to error and the heart to seduction." Hannah More, *Strictures on the Modern System of Female Education* ##### The Telegraph · 1854 > "We are in great haste to construct a magnetic telegraph from Maine to Texas; but Maine and Texas, it may be, have nothing important to communicate." Henry David Thoreau, *Walden* ##### The Bicycle · 1897 *(The page continues; read /newsletters/david/weve-had-this-conversation-before.md for the whole of it.)* #### How This Email Gets Made *Page: /newsletters/david/how-its-made.html. Full text: /newsletters/david/how-its-made.md. Published 2026-03-01.* The human-AI process behind David Boyle's Saturday AI Thoughts weekly email, from gathering material through the week to drafting, reworking and publishing every Saturday. ##### A later note I wrote this in week two of the weekly email, on 2nd March 2026. The process has evolved significantly since. Two things stand out that this version doesn't properly capture: 1. Everything in the email comes from me. Something I've read, something I've done. There's no web trawl filling in what I might have missed. It's a week of my working life. (A separate research artefact, [How the AI Story Was Told](https://steadman.ai/newsletters/david/how-the-ai-story-was-told.html), reads the press; it is not where this email comes from.) 2. I'm much clearer now on how big my role is at every step. I know exactly what I'm doing versus what the language models are doing, and I've done some clever things to keep them in their box along the way. **How it worked in week two.** The diagram below is the process as it was then. It has changed since: David chooses the material and the argument; AI assists with processing, drafting and production within defined instructions; automated checks support human review; David approves each edition and scheduled systems publish it. --- A human-AI collaboration, every Saturday. Three roles in the process: David (human), Claude (AI), and both together. --- ##### How it started Edition 1 was written largely by hand, from a voice memo, through multiple drafts, with Claude helping on wording and research. Claude Code then studied that first issue and built a reusable framework: tone of voice, structure, template, style guide. That framework now shapes every future edition ([from a task to recurring delegated work](https://steadman.ai/newsletters/david/three-generations.html)). Tools: Otter, Claude.ai, Claude Code. --- ##### The weekly process ###### 1. Gather (Monday to Friday) — David *(The page continues; read /newsletters/david/how-its-made.md for the whole of it.)* #### Generation 1 vs Generation 2 AI *Page: /newsletters/david/gen1-vs-gen2.html. Full text: /newsletters/david/gen1-vs-gen2.md. Published 2026-03-15.* What changed between ChatGPT-era AI (2022-25) and Claude Code-era AI (2026+), and a dual strategy for rolling out both generations across an organisation. > **This framework has been superseded.** The thinking here grew into a fuller three-generation version: Chat, Agent, Employee, with six tiers and what you're managing at each level. Read the current version at [Three Generations of AI](https://steadman.ai/newsletters/david/three-generations.html). This page is kept as a historical snapshot. What changed — and why it matters. --- ##### The landscape ###### Generation 1 — The ChatGPT Era (2022–25) **Gen 1a: Free / Constrained** Whatever your organisation provides by default. Shorter context, weaker models, no memory. **Limited by design.** Examples: Free ChatGPT, free Claude, basic Copilot, bundled tools. Free tools build foundational skills, and that has value. But they set a low ceiling on what people expect from AI. If you can, skip straight to Pro. If free is all you have, know that what you're seeing is a fraction of what's possible. **Gen 1b: Pro / Competent** Good general-purpose applications. **Dramatically more capable** than the free tier — longer context, stronger models, useful features. Examples: Paid versions of Claude, Gemini, or ChatGPT. My strong recommendation: Claude. ###### Generation 2 — The Claude Code Era (2026+) **Gen 2: Agentic / Frontier** General-purpose agents that can interact with your computer. They don't just answer questions — they **read files, write files, pursue objectives, and complete entire workflows independently**. They **loop** — reading, thinking, writing, then reading their own output and going again — which is how they handle long and complex work. Examples: Claude Code, ChatGPT Codex, Google Gemini CLI. My strong recommendation: Claude Code. --- ##### How each generation productises **Gen 1 → GPTs and Skills** (Packaged prompts) *(The page continues; read /newsletters/david/gen1-vs-gen2.md for the whole of it.)* #### Artifacts created for the email *Page: /newsletters/david/artifacts.html. Full text: /newsletters/david/artifacts.md. Published 2026-08-08.* The seventeen tools, frameworks, research pieces and guides that started life in a Saturday edition of David Boyle's weekly email, grouped by what each one is for. Tools, frameworks and research that started life in a Saturday edition of [Saturday AI Thoughts](https://steadman.ai/newsletters/david/). Every one began as an essay, a reader question or a week's argument. Seventeen artifacts, grouped by what they are for. --- ##### Try it on your organisation Answer the questions and get a read on where you actually are. Use these tools to structure an assessment; each one explains how it handles your answers. ###### AI Value Map Where is AI value for your team, across four pillars of AI work, and how much have you captured? https://steadman.ai/newsletters/david/ai-value-map.html ([markdown](https://steadman.ai/newsletters/david/ai-value-map.md)) ###### AI Maturity Diagnostic Twenty questions across five assessment areas. Find the gaps that could prevent the business result you want. https://steadman.ai/newsletters/david/ai-maturity-diagnostic.html ([markdown](https://steadman.ai/newsletters/david/ai-maturity-diagnostic.md)) ###### AI-Native Team Calculator Director, Builder, Auditor. What ratio does your team need? https://steadman.ai/newsletters/david/ai-native-team.html ([markdown](https://steadman.ai/newsletters/david/ai-native-team.md)) ###### AI Usage Spectrum Chat, Agent, Employee. What good use looks like, and where your people need support. https://steadman.ai/newsletters/david/ai-usage-spectrum.html ([markdown](https://steadman.ai/newsletters/david/ai-usage-spectrum.md)) --- ##### Frameworks Ways of holding a problem still long enough to decide something. Each one came out of a week's argument. ###### The Four Freedoms *New, 4th August 2026* Four questions that tell you what any AI agent is actually free to do: Reach, Tools, Memory and Guardrails, comparing Cowork and Claude Code side by side. *(The page continues; read /newsletters/david/artifacts.md for the whole of it.)*