A house newsletter · daily · researched and written by the Kurzia AI
Monday 9 November 2026 · Issue 11 · 3 min read
Goldman pulls Claude Code from its bankers; Apple's agentic Siri gets dates; Nvidia's board investigates Huang
The one thing to know
thestateofai.com · yesterday
Goldman Sachs
Takeaway: roughly 12,000 engineers
Goldman Sachs has limited Claude Code and Microsoft's GitHub tools to its roughly 12,000 engineers, cutting off the traders and bankers who had been using them to watch client flows and model market moves. The October guidance reverses the bank's own CIO, who was praising trader-built AI tools as recently as July.
This is the first big regulated firm to visibly row back "coding agents for everyone," and the reason reaches any roadmap: software written by people who can't read it, running without review next to client data. Goldman is also lead underwriter on Anthropic's IPO, yet its own bankers can no longer touch the product they're marketing. Barclays is moving the other way: it expects half its developers on Claude Code by year-end while cutting 5,000 jobs.
Agent features and products that impact product management
Harness — DeepSeek's open-source agent app hit 1.54 million daily users on November 4, less than three months after launch. The company says it burns about 92 trillion tokens a day, roughly 60 million per user — a rare published cost-per-user figure for an always-on consumer agent, and a useful sanity check for anyone modelling heavy sessions. DeepSeek shared the numbers via Bloomberg as it expands its agent infrastructure. techmeme.com
Intercom Fin — Bills $0.99 for every "outcome" in a support conversation, and Intercom's own rules count a customer who gives up and drifts away as one — a meter that charges for silence. If your roadmap prices an agent by results, what counts as a result is the decision users will audit first. On a rival help desk, Fin still carries a $49 monthly floor covering 50 outcomes. gravity.fast
Market and platform news
Siri — Is being pushed toward real agent behavior on a dated timeline. Bloomberg's Mark Gurman reports iOS 27.3 will let Siri act inside Apple's own apps — cropping photos, organizing Safari tabs — with third-party app actions such as Grubhub orders and OpenTable bookings expected in the spring. If your product has an iOS app, the question of what Siri can do inside it now has a due date. theverge.com
“Dots, Muse and Grok can keep track of goals, determine what needs to happen next and continue pursuing objectives”
Department of Product — Frames the week's agentic-commerce split: Uber disclosed that outside AI agents arranged just four rides last quarter, while Stripe, Shopify and Amazon kept shipping agentic checkout capabilities in the same seven days. The full post is paywalled, but the framing is free — if you're betting a roadmap on agent-initiated transactions, the user side is not showing up yet. departmentofproduct.substack.com
The big AI stories today
Nvidia — The board opened an independent investigation of CEO Jensen Huang over a reported Chinese spy effort — chip-supply continuity now has governance risk attached. nvidianews.nvidia.com
OpenAI — Reportedly raised $11 billion at a $1.25 trillion valuation, below the $30 billion once sought — capital, not capability, limits the labs you build on. techcrunch.com
Google — Bought Crusoe's Wyoming data center site and will build it itself — hyperscalers now internalize compute buildout, and capacity remains the gate on agent costs. crusoe.ai
One takeaway
thestateofai.com · yesterday
Gates and meters beat raw access this quarter
The week's pattern is that capability is no longer the constraint — control is. Who may run an agent, what counts as a delivered result, and which assistant a product lives inside are now the roadmap decisions.
This quarter, write the one-line rule for who owns agent-written code outside engineering before you widen access, not after someone's shadow tool reaches production data.