David's Saturday AI Thoughts

Kids these days

WHAT HAPPENED THIS WEEK * AI displacement now shows up in the US government data at both ends of the career ladder: A Bloomberg analysis of new BLS figures finds every one of the eighteen occupations the BLS classifies as AI-exposed has lost jobs over the past year, even as US payrolls grew 0.8% overall. Customer service representatives shed 130,180 jobs, 4.8% in a single year; interpreters down 24% over three years; credit authorizers down 26%. The exception that confirms the rule: medical secretaries up 15.8%, the cluster that needs a body in the room. The same picture shows up at the other end of the funnel. The Economist this month plotted US graduate full-time employment against AI exposure: computer science and information sciences graduates are down 10 to 15 percentage points since 2022; philosophy and psychology graduates held steady or gained. The displacement isn't just to the people already doing those jobs. It's to the people trying to start in them, and what they should be studying may not be obvious to anyone yet — a thread the essay returns to via Elliott's homework. * The UK's data regulator has put AI hiring tools on formal notice. Sixteen organisations have already had a letter: The Information Commissioner's Office issued formal guidance this week saying that AI-driven CV screening, candidate ranking, and video interview analysis without "meaningful human involvement at every consequential stage" may already breach UK data protection law. Sixteen organisations have been written to directly. The consultation closes on 29th May, six days after the edition lands. A concrete Monday-morning task for any leader running a hiring pipeline: get the full list of AI tools in use across the funnel, decide which involvements count as "meaningful" against the ICO's test, and put a response into the consultation. The window is genuinely short. * Salesforce will spend close to $300 million with Anthropic this year. Marc Benioff says the engineering productivity gains made it the easiest line in the budget: Marc Benioff disclosed that Salesforce is on track to spend close to $300 million with Anthropic over 2026, with most of the spend on coding, justified by engineering productivity gains of more than 30%. Separately, Anthropic announced a $200 million partnership with the Gates Foundation focused on global health. A Fortune 100 chief executive treating the model layer as a procurement line item, not a research expense. The bigger question is who in your firm is allowed to commit that kind of capital, against what kind of evidence, and how quickly. WHAT TO TRY * When the output goes wrong, shrink the task: Justin Skycak put it as a principle for skill acquisition this week: shrink the unit of practice until the mistake has nowhere to hide. The same rule applies to working with language models. Sprawling prompts produce sprawling failures you can't diagnose. Break the task into its smallest meaningful unit, run it, inspect the output, then rebuild. If you can't immediately see where it went wrong, your chunk is still too large. * Ask AI questions it can't possibly know the answer to: A marketing lead at a global firm told David this week she's running a five-minute stress-test on every AI tool she's thinking of trusting. She uploads her own data, asks the model to use only that data, then asks it questions she knows the data can't answer. Some models fabricate regardless ("53% of women in the northeast states feel..."). She's learned what its confident-but-wrong mode looks like before depending on it for an answer she can't independently check. Worth doing once on every tool you rely on. * Run your day past AI before you start it: A senior leader described her commute habit to David this week. She opens Claude, asks it to review her calendar and her email, then asks it to surface what she needs to read before each meeting, what's carrying over from yesterday, and which emails in her inbox need replies before the day eats her. Five minutes on the train, and the day is scoped from outside her own head. "Just a nice little daily habit," she said. Try it tomorrow. Read the full edition with all links and sources [https://steadman.ai/newsletters/david/#edition-2026-05-23]

What boards accept

WHAT HAPPENED THIS WEEK * The METR capability curve just went from one hour to one day. The unit of AI autonomy is now measured in human work-days, and the doubling holds on a log scale: METR, an AI evaluation lab, measures how long an autonomous task an AI can complete reliably. In early 2024 the answer was minutes. In May 2026, with Anthropic's forthcoming Claude Mythos Preview, it's sixteen hours of work a human would have done. The number isn't the headline; the curve is. From minutes to a day in twenty-four months, on a log scale that had previously held steady. If the next two years look like the last two, the unit becomes a week, then a month, then a year of human work at the press of a button. Plan accordingly. * Half of organisations have already redesigned core workflows around AI, and a fifth have built new business models. The gap framing misses the story: BCG's AI at Work 2025 survey of 10,635 employees across eleven countries reports that 72% of organisations are running generative AI tools, 50% claim to have redesigned end-to-end workflows around them, and 22% claim to have built new business models on top of them. Three and a half years after ChatGPT launched, a fifth of firms claim new business models because of AI. Half have rewired core workflows. Syed Ijlal Hussain, who surfaced the chart on X, framed it as a gap. David flips it: the 22% are doing what most boards he's working with haven't started. * Anthropic just passed OpenAI in US business AI spend. The strategy lesson is older than AI: pick an audience and serve them: Ramp's AI Index, built from anonymised spend data across its US business customers, shows Anthropic taking 34% of paid AI subscriptions in its May release, ahead of OpenAI on 32%. The first crossover. Anthropic's share has roughly quadrupled in a year. David's read: this was inevitable from early on. Anthropic stayed fixated on the enterprise user while OpenAI chased every consumer headline. Slow perseverance against a chosen audience won. The lesson isn't really about AI. Pick an audience. Set your strategy around their needs. Keep your head down and serve the people you said you'd serve. WHAT TO TRY * Hand over the context, not just the question: Experienced leaders have context and are short on time. AI tools convert context into time saved, but only if you hand the context over. A leader David spoke with this week made the leap from asking to delegating. "Here are the Q2 numbers, last quarter's board paper, and the three things the board flagged. Draft it." Same model, same minute. Asking gets you an outline. Delegating gets you a draft. * Build a personal skill, and add a rule to it every Sunday: A person David worked with this week reviews 100-page reports from their team on Sunday nights — typos, inconsistent language, logic gaps. A simple review skill in Claude now catches them. The compounding version is their own preferences layered on top: this market "is expected to grow" not "will grow"; never "dropping precipitously." Every time they catch a miss the model didn't, they open Claude and say "add this rule to my personal review skill." Whatever skill you build, the compounding habit is the same. Catch what the model missed. Add the rule. Trust the skill more next week. * Schedule a daily AI briefing. The use cases will follow: AI tools sit closed until you open them. That's a real reason senior leaders bounce off: not bad prompts, but a tool that requires you to think of the use case first. Scheduled tasks invert this. Every morning at three, David's reads his inbox from the previous day, prepares one-paragraph briefings for every meeting on the day's calendar, and emails a short summary that takes three minutes to read with breakfast. The tool changes from something you open to something that opens your day. Read the full edition with all links and sources [https://steadman.ai/newsletters/david/#edition-2026-05-16]

Choosing is the work

WHAT HAPPENED THIS WEEK * AI-adopting firms are growing headcount, not cutting it: A Goldman Sachs analysis circulating this week, charted by Callum Williams of The Economist, shows US firms that have adopted AI report net positive employment growth across the past six months. Finance, insurance, arts and entertainment sit at the positive end; transportation and food service show modest decreases; the all-industries balance is positive. Self-reported and narrow-window, but the direction matters. David reads it as Jevons playing out: AI-adopting firms grow because productivity gains expand what their people can do faster than they replace them. The risk is falling behind the firms whose people are using AI to scale. * Five percent, not fifty: the candid private-equity number: Pete Stavros, co-head of global private equity at KKR, told the Milken Institute conference last week that AI is improving portfolio company earnings by about 5%, not the 50% that a revolutionary technology should offer. Five percent across a portfolio of billions is real money. It is also a long way from the scale of growth many hope for. The gap between what feels possible and what the spreadsheets show says something about where the bottleneck actually sits, and these days it isn't the AI. Sets up the essay's punchline that choosing well, not building harder, is what's left. * Both AI labs went into private equity the same day: On Monday, Anthropic announced a $1.5 billion vehicle with Blackstone, Goldman Sachs and Hellman & Friedman. Engineers from Anthropic will embed inside the consortium's mid-market portfolio companies and build custom Claude workflows. The same day, OpenAI finalised a $10 billion joint venture; Google is reportedly in talks to do the same. What was reported as "AI labs raise more money" is actually a category change: the AI labs are turning into distribution-led services firms with a model attached, lifting the forward-deployed engineer pattern from Palantir. David's worry: PE has run the same cut-first playbook for forty years; the bigger unlock from AI is the opposite move, people making sharper decisions with a tool thinking alongside them. That value compounds; what a forward-deployed engineer hard-codes into a workflow does not. WHAT TO TRY * Push it harder, then skillify, then iterate: The simplest workflow upgrade David has coached this year, and it stacks. Push it harder: when the model gives you a perfectly reasonable answer, tell it to do more searches and use credible primary sources, government statistics, peer-reviewed papers. Three extra words, often. The next round does real research and surfaces what the first pass skipped. Skillify it: when a session produces something useful, ask the AI to turn the exchange into a reusable skill, a folder with a description and a prompt any future session can pick up; encode the mistakes as guardrails. Iterate: when the skill misbehaves, just say "change yourself so X" and it rewrites itself in place. Skills become living artefacts, not fixed ones. Works particularly well in Claude where skills are first-class. * Shadow your most AI-pilled employee for two days: Matt Stockton, an operator and investor, made the case this week. Find the rabbit-holed colleague (not the keenest, not the head of digital transformation, the one who already knows what they're doing) and watch how they actually work for two days, then do the same work yourself for a week. "AI lives on command lines right now, it does not live in PowerPoint jargon slides." Most senior leaders consume AI through polished demos and vendor decks, then make strategy decisions with no felt sense of the technology's real constraints. Cheapest fix going. Questions to ask while shadowing: which tools and settings are you using, show me the actual prompt you typed, what did the model get wrong, what did you stop doing manually three months ago that's still in my diary. * Ask AI to build you an HTML slide deck instead of PowerPoint: Just ask. "Build me a slide deck on [topic] in a single HTML file." Most chat tools can do this now. Why bother? PowerPoint output from AI is uneven, with broken chevrons, mis-rendered shapes, and two hours of touch-up. HTML slides come back closer to ready, are infinitely easier to edit (you talk to the file rather than wrestle with a template), and unlock things PowerPoint can't do: an inline edit-mode toggle that turns text into contenteditable, a speaker-view page with notes and a timer, narrate buttons that read slide notes aloud, a Q&A page that answers from the deck content only. Single-file decks travel better than multi-file projects: you can email it, host it on any static page, screen-share it without screen-sharing software. Read the full edition with all links and sources [https://steadman.ai/newsletters/david/#edition-2026-05-09]

The bill and the harness

David builds the case that flat-rate AI pricing is dying and that the buyer's question is no longer 'how much will this cost' but 'where does the spending compound'. He opens at a Las Vegas buffet that closed on 31st May, then moves to the supplier-side news: three of the four biggest AI vendors switched pricing in the last few weeks (Anthropic stripped bundled tokens out of Enterprise seats in mid-April, OpenAI took Codex pay-as-you-go a fortnight earlier, GitHub moves every Copilot plan to usage-based billing on 1st June, and an Anthropic manager admitted Pro and Max tiers have been outgrown). He brings in two friends' worried voice notes from the buyer side: a friend in Tokyo asking what happens when bills go up five or ten times, and a partner at a professional services firm naming the outsourcing trap. He explains the supplier maths (unit prices falling roughly tenfold a year, his $200-a-month Max plan delivering $500 a day of equivalent API use, unsustainable) and the buyer maths (Jevons Paradox: cheaper energy made coal use rise, not fall). The radiologist is the modern Jevons: Hinton's 2016 'stop training radiologists' was right about the models and wrong about the radiologists. Ten years on the US has six thousand more of them and pay is up roughly seventy per cent. Punchline: the bill rises either way, the question is whether the spending compounds in the model (a utility cost) or in the harness, the layer of instructions, context and workflows that wraps the model (an asset nobody else can buy). Intercom doubled engineering velocity in nine months on exactly that bet. WHAT HAPPENED THIS WEEK * AI adoption stalls one layer below the executive sponsor, at the line manager: Gallup data (Q4 2025) finds AI use correlates more strongly with managerial endorsement than with tool access. In firms where the manager actively supports AI, 80% of staff use it weekly; where they don't, that drops to 44%. In the public sector: 65% versus 37%. Procurement and licences are the easy part. Adoption lives or dies one layer below the top. * The frontier-model leaderboard is now refreshing in weeks, not quarters: The Epoch Capabilities Index now shows GPT-5.5 Pro and Gemini 3.1 Pro above 155, up from GPT-4o's 128 in mid-2024. Seventeen frontier releases compressed into under two years with no visible plateau. Epoch researcher Greg Burnham: 'I don't know when Opus 8.2 will be shipped, but GPT-9.1 will be shipped that afternoon.' Build workflows and judgment around capabilities, not named models. * Six VC firms, one investment thesis: Linas Beliunas read the published 2026 investment theses of six of the biggest venture firms side by side and found the same handful of AI bets in all of them: AI-native enterprise software, AI agents for physical and industrial work, vertical AI software in legal, finance, healthcare and construction, multi-agent orchestration, and health AI. His sharper conclusion: the most valuable software companies of the next decade won't look like software companies. They'll look like law firms, factories and hedge funds run by teams of ten. WHAT TO TRY * Pick one tool, get fluent, then refine your harness: A leader David spoke to had spent weeks running the same task through ChatGPT and Claude side by side, then asking each to review the other. Genuinely interesting and occasionally useful, but they couldn't decide which to use when. The advice from someone who has been at this longer: pick one, become very good with it, learn its quirks, then build the layer above it that reflects how you actually work. Most of the value is in fluency with one tool, not coverage across two. * Force yourself to change something on every AI output before you ship it: Came up at a senior training session this week, as the room debated when the 'check, edit, own' model breaks down. Increasingly the edit feels optional because the output looks good. Whatever the AI gives you, a paragraph, a slide, a summary, add something, remove something, or reorder something before it leaves your hands. The rule of 'always change something' is a forcing function against brain rot. * Skip the slides, build the page: In a senior strategy session this week, the most-praised artefact in the room was not a deck. It was a web page someone had built to walk teams through their thinking. Building it took a clear thought, a text file, and a small Claude skill they had set up once. Their harness paying off in front of the room. Read the full edition with all links and sources [https://steadman.ai/newsletters/david/#edition-2026-05-02]

Rise of the auditors

AI-native teams need three roles: Director, Builder, Auditor. Execution is cheap, verification is expensive. Most organisations have zero Auditors and are shipping nothing because nobody is named to check. WHAT HAPPENED THIS WEEK * <10% of organisations scale agents beyond pilots (McKinsey) * GitHub paused Copilot signups / Uber blew 2026 AI budget / Goldman inference costs approaching headcount parity * 29% of employees sabotaging AI initiatives (Writer survey) WHAT TO TRY * Ask is this the simplest version (Cantrill laziness) * Audit cold: different model, fresh context * Save one reusable AI workflow (Chrome Skills) Read the full edition with all links and sources [https://steadman.ai/newsletters/david/#edition-2026-04-25]