Goldman pulls coding agents from its bankers; Nvidia's board investigates its own CEO

🔥Today's one thing

🔍What is real, what is not

• October jobs data — US professional services — consulting, law, accounting, the white-collar work said to be most exposed to AI — added 70,000 jobs in October, the sector's best month since July 2022, the Wall Street Journal reports. The wipeout boards keep hearing about is still not visible in hiring; plan headcount on your own numbers, not the fear. Wall Street Journal via Techmeme

• Harness — Chinese AI firm DeepSeek published adoption figures for Harness, its free, open agent app: 1.54 million daily users on November 4, under three months after launch. Consumer-scale agent demand is real — but these are the vendor's own figures, released by DeepSeek and not independently checked, so treat them as a marketing data point. The company says the app burns about 92 trillion tokens a day, roughly 60 million per user — a usage level that helps explain why vendors keep admitting agents are expensive to run. Bloomberg via Techmeme

📈Money and the market

• Siri — Apple is racing to make Siri handle multi-step jobs on a user's behalf, with CEO John Ternus pressing software chief Craig Federighi to move faster after concluding Apple is behind OpenAI and Meta. If a genuinely capable Siri ships on a billion iPhones, it decides which services agents reach first — a threat and a channel for anyone selling to consumers. Per Bloomberg's Mark Gurman, deeper control of Apple's own apps arrives with iOS 27.3, and third-party actions such as placing a Grubhub order or booking OpenTable are expected in the spring. The Verge

• Wyoming data-centre site — Google has bought the southeast Wyoming site Crusoe spent two years preparing and will build the campus itself, after Crusoe reportedly failed to sign customers and Google raised concerns about cost and timing. For anyone renting AI capacity, the build-out is concentrating further in a few very large hands. Google has committed $17 billion to developing the campus; the price paid to Crusoe was not disclosed. Crusoe raised $3.9 billion at a $30.9 billion valuation in September despite the setback. Crusoe

• OnePWR generator deal — Anthropic bought commitments for 1.7 gigawatts of gas-fired generators from equipment supplier OnePWR for $3.4 billion, the Financial Times reported citing sources — described as the largest purchase of its kind known from an AI lab, made by going around the usual suppliers to grab scarce power equipment. Capital spending like this ends up in the prices customers pay. The report is sourced; neither company has confirmed the terms. Financial Times via Techmeme

• Anthropic's listing price talk slides toward $1.1 trillion. The mood among tech investors online is blunt: many see no clear growth story at prices near Meta's. A weak debut would toughen every AI vendor's next sales negotiation — good leverage for customers. Commenters on the thread compared the price to Meta's $1.55 trillion market value and Apple's $5 trillion. bloomberg.com · from the Hacker News discussion

  “literally what is the upside? They're supposedly worth as much as meta despite having nowhere near the customer base or years of work?”

⚠️Risk the board will raise

• Nvidia board investigation — Nvidia's board has opened an independent investigation — a committee of outside directors with external counsel — into reports that CEO Jensen Huang was targeted by a Chinese intelligence operation. Huang has admitted an affair with the executive at the centre of the reports and apologised to staff. The board moved within a day and says it sees no sign secrets were compromised: a template for handling a reputational hit fast. Three House committee chairs have summoned Huang to a closed-door briefing on November 18; his companion's employer, Nvidia customer Megaspeed, is separately under US investigation over chip smuggling. NVIDIA Newsroom

  “The Board prioritizes U.S. national and economic security, which is paramount to the interests of NVIDIA and its shareholders.”

🌍The big AI stories today

• CAISSI — Commerce's AI-safety centre was reportedly asked to report informally to a White House office — a test of whether US oversight of agents stays independent. Associated Press via Techmeme

• H200 export licences — Two Chinese firms cleared to buy restricted Nvidia chips keep empty offices and have executives tied to sanctioned military suppliers; chip-export vetting is thinner than claimed. Tom's Hardware

• Claude outage — Anthropic's Claude returned elevated errors for 14 minutes yesterday across its app, API and coding agent — brief, but a reminder to name a fallback vendor. Anthropic status page

💡One takeaway