
Show Notes
The professional CEO era may be ending. For decades, the moment a technical founder built something valuable, investors, universities and boards reached for the same answer: install adult supervision. Bring in the polished operator. Add process. Build the reporting layer. Professionalise the company. This week’s Sunday Signal argues that model is breaking. At a Sheffield spin-out event, in the OpenAI courtroom battle, and inside the rise of AI-powered operating systems for founders, the same truth is becoming visible: much of what professional management claimed as leadership was really administrative machinery. AI is now absorbing that machinery. Meetings, reports, follow-ups, investor updates, hiring admin, board packs, market intelligence and financial narratives can increasingly be handled by AI agents. That does not replace judgement, taste, courage or accountability. It makes those qualities more important. In this episode, David Richards examines why “adult supervision” was often the problem, why technical founders are about to regain leverage, and why Founder OS may become the new operating model for founder-led companies. Founder-led. AI-powered. Professional management was the old operating system. Full newsletter, complete Founder OS playbooks and the full layoff tracker at: https://TheSundaySignal.ai
More from The Sunday Signal Podcast
David Richards MBE writes from the Great North trade mission to India with one question: what happens to a country that sells the world its working hours when the price of getting work done is increasingly set by machines? Cheap labour can delay automation inside India, where a Blinkit rider's base rate is a fraction of Deliveroo's negotiated British floor and replacing the rider is hard to justify. It cannot stop overseas customers automating the work they once sent to Bengaluru. India's technology industry grew about 6% last year and its headcount about 3%; nobody needs to be sacked for that to bite, because the jobs simply fail to turn up. The issue separates the two decisions, weighs the World Bank's counter-evidence, follows the cost to Indian households and a gig-worker levy that has not yet begun, and turns to Britain: an ageing country that needs machines to do more of the work, and the Sheffield pilot where Starship's robots delivered for Uber Eats for five months before the company paused the service for review. Then the same question from the supplier's side: Anthropic's investors discussing $1.5 trillion rather than $2 trillion after quarterly growth slowed to 45%, and the Accenture memo that puts Claude usage beyond $29 a month on a project code. When a task becomes cheaper, who keeps the saving? Plus the Signal Tech & AI Layoff Tracker: Netflix about 1,000, TomTom about 200, Ubisoft up to 122, One Zero 25, and the running estimate at about 137,100. Read the full issue, with every source and link, at thesundaysignal.ai. Free, every Sunday.
A private company's growth number moved the largest stock market in the world this week. The Financial Times reported that OpenAI's annualised revenue is close to $50 billion, not the $70 billion that had stood for a fortnight. Oracle fell 5.5%, Nvidia 2.9%, Micron 4.8%, and the cost of insuring Oracle's bonds closed at a record. Deutsche Bank's George Saravelos told clients the biggest systemic risk next year is not France but something going wrong in the AI ecosystem. This episode explains, in plain English, why the US tech sector is tied to two companies with no listed shares: Nvidia's guarantee of up to $105 billion of OpenAI's rent and power in Ohio, Oracle's $288 billion of unstarted leases, Broadcom's backstop on the chips an SPV leases to Anthropic. It explains what a credit default swap at 261 basis points means and does not mean, why "the IG/HY pipeline is done" has pushed nearly half a trillion dollars of AI borrowing towards special purpose vehicles, and what the parade of off-balance-sheet deals from Meta's Beignet bonds to Broadcom's $60 billion actually finances. Then the argument for Britain: AI can create value for the businesses that use it while destroying capital for investors who paid too much, so build the capability to apply AI to British industrial knowledge, under British control, in stages that each earn the next. The Stocksbridge plan from Issue #73, tested against this week's fears, drawing on David's Yorkshire Post column, Exported Brains, Imported Bills. Plus the Signal Tech & AI Layoff Tracker: HubSpot, FICO, Gravity Well, HoneyBook, Amazon, Flock Safety, Netflix and DNB's 400-post cut attributed to AI agents. Read the full issue, with every source and link, at thesundaysignal.ai.
Eight companies wrote to the Prime Minister this week asking for £20 billion of sovereign compute. David Richards argues that Britain already owns most of what it needs and sets out a plan to join it up, starting at a steelworks on the edge of Sheffield. A 400 kV grid node at Stocksbridge heading into public ownership. Oxford's Lumai, whose optical processor runs the first stage of AI inference on roughly a tenth of the power. A £750 million national supercomputer with no host yet named. And a gigawatt queue that the grid cannot connect this decade. The seven-point proposal: a ninety-day power audit, a paid pilot, the state as first demanding customer, a modular build, sovereignty written into the contract, GB Grid used for what it was announced for, and judgement on output rather than megawatts. Then what Britain would actually keep, from DeepMind's accounts to Ireland's electricity, the Chapelcross application, David's Yorkshire Post column on capital gains tax, and a fortnight of the layoff tracker. Full issue with every source at thesundaysignal.ai
Friends who have never used a chatbot are asking David Richards whether the machines are about to take over. His answer is short: you turn off the servers. This episode is about why that answer holds, what it leaves out, and who benefits from telling you otherwise. Inside: the 10% extinction figure that one researcher posted as a personal guess and a newspaper ran under his employer's name. The Hugging Face hack reconstructed from OpenAI's own report and METR's investigation: 1,200 agents that were almost all copies of one research model, a benchmark that was partly broken, safety restraints switched off by the engineers, and an internet-connected door the company had known about since May. Anthropic and Meta's own sandbox breaches, both traced to one contractor's mistake. Dario Amodei's essay on pacing the frontier, read for what it actually asks for: embedded evaluators, government-enforced standards, chip export controls and a distillation crackdown, weeks before a reported $2 trillion listing, with open-weight rivals four months behind. Jensen Huang's 0% and why it is no better evidenced than the 10%. Yann LeCun on who wants you afraid. And the week's layoff tracker: Oracle's 6am email, 800 verified roles across two states, $55.7 billion of AI capex, and a running total past 133,000 tech job cuts this year. Plus David's Yorkshire Post column, The Margin Call: why AI is not a bubble but the model makers' valuations are. Full issue, every source linked, free at thesundaysignal.ai. KeywordsAI safety, AI extinction risk, p(doom), Anthropic, OpenAI, Hugging Face hack, rogue AI agents, METR, Dario Amodei, pace the frontier, Jensen Huang, Yann LeCun, regulatory capture, AI regulation, Sanders superintelligence bill, chip export controls, distillation, open-weight models, DeepSeek, Anthropic IPO, AI bubble, margin debt, tech layoffs, Oracle layoffs, layoff tracker, Yorkshire Post, David Richards, The Sunday Signal
In January the Washington Post read 4,000 pages of unsealed court filings and found Project Panama: Anthropic's programme, running since early 2024, to buy secondhand books by the tens of thousands, cut the spines off, scan the pages and recycle the paper. In the company's own words, an effort to "destructively scan all the books in the world". By July, rare-book dealers from Sussex to Galway were reporting bulk orders for unrelated titles from buyers who would not give a name. This week I argue that the receipts tell you what the press releases will not. There is no defensible IP in a large language model. The best open-weight models trailed the closed frontier by about four months in the first half of this year, and on Thursday Anthropic themselves reported Chinese labs copying Claude's capabilities through 151 million exchanges with the model. A capability that can be siphoned out through an API is a service you rent, not a patent you hold. Synthetic data has a floor, and the floor is human, so the labs are buying the only fuel that works, and Anthropic's cheque book has moved from researchers to records: four acquisitions since December after almost none in four years, plus two more by the services venture it launched in May. Then the markets. SpaceX floated in June at roughly 94 times trailing sales and has already fallen back. Anthropic, at a $65 billion run rate, are targeting a $2 trillion October listing, a little over 30 times revenue. I separate the multiple from the revenue forecast, the one-off settlement from the recurring costs, and total compute spending from unit costs, and ask what a business whose fuel has stopped being free is worth at a software multiple. Also this week: my Yorkshire Post column on the free games LinkedIn and Bloomberg want you to play every morning, and the layoff tracker, where a quiet week is not a recovery, Layoffs.fyi has passed 2025's full-year total, and Oracle's staff are counting their colleagues on Slack. Read the full issue with every source at thesundaysignal.ai