A house newsletter · daily · researched and written by the Kurzia AI
Wednesday 11 November 2026 · Issue 13 · 3 min read
Anthropic prices at up to $1.4T; Amazon cuts one AI task's cost 98%; agent shopping stays tiny
Today's one thing
TechCrunch · yesterday
Anthropic IPO
Takeaway: $1.404 trillion
At the top of the range it would be worth $1.404 trillion — up from $965 billion in May, but far under the roughly $2 trillion it was reported to want. The filing shows the strain behind it: $33.2 billion of nine-month revenue against a $104 billion net loss.
This is the reference price for the whole AI supply base. A public Anthropic must defend that number every quarter, so expect sharper pricing and heavier sales pressure from every vendor you buy from. Alongside the sale, AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC are buying $26 billion of non-voting stock, Google and Nvidia at $10 billion each — Nvidia's slice is about a tenth of all the new money Anthropic expects to raise this month.
What is real, what is not
Strands Decider — An internal Amazon team replaced the big AI model that sorted its tickets with a much smaller one that only picks from a fixed list — quality held, and the cost of labelling a million tickets fell from about $22,200 to roughly $500. Amazon's own numbers on its own product: proof of possibility, not a promise. But it marks where the money is — high-volume internal chores, not showy customer features. The team ran both models on live tickets before switching; going back was a 45-second change. Strands Agents blog
“Running both on real traffic, then judging every disagreement, told us more than any offline eval would have.”
Agents still aren't ordering the dinner. Uber and DoorDash told investors that orders placed by AI agents remain a rounding error: about 500 Uber Eats orders and four rides in what Uber called one of its biggest weeks, and only a few hundred total for DoorDash. For anyone budgeting for agentic-shopping revenue, the customers are not there yet; the platforms are still building for demand that has not arrived. Uber's 320 million trips in the last week of October make the agent share effectively zero. Business Insider
“It is very, very early, and very limited traction that we have seen so far.”
OpenAI's leaked numbers: fast growth, still huge losses. Documents reported by the Wall Street Journal, and not confirmed by the company, show OpenAI's revenue jumped roughly 50% in a quarter to about $10 billion, while losses narrowed but stayed enormous. Unverified, but they frame the sector's economics: even the market leader spends far beyond its income, which is why pricing pressure keeps landing on customers. The adjusted operating margin moved from -150% in Q2 to -109% in Q3, per the leaked documents. Techmeme / Wall Street Journal
Money and the market
Imagine Enterprise — A joint Microsoft–IBM platform selling pre-built workplace AI agents plus consultants who install them inside your company, aimed first at financial services, insurance and retail. Two incumbents bundling agents with consulting target companies that would rather buy than build; expect it to surface in your next Microsoft renewal conversation. Gartner finds only 4% of CIOs have plans to adjust roles for a workforce that includes agents — the organisational side lags far behind the buying. CIO Dive
CoreWeave — The AI data-centre landlord beat expectations with quarterly revenue of $3.66 billion, up 168% year over year, and raised its 2026 outlook — but still lost $709 million in the quarter. The compute under every agent plan is still growing fast and being built partly on borrowed money; check the solvency behind capacity promises before signing long contracts. Its backlog of signed contracts reached $132.5 billion, up from $104 billion in June. Techmeme / CNBC
Claude Meetings — Anthropic's push into everyday office work: Claude joins video calls as a visible participant, drafting plans and answering questions while people talk, with shared whiteboards and spreadsheets launched alongside. Anthropic, OpenAI, Microsoft, Google and Meta are all converging on the office suite — decide now which vendor ends up holding your meeting records and documents. It launches in test form on Team and Enterprise plans with Google Meet, and Claude names itself in the chat and leaves automatically if an outside guest joins. TechCrunch
eBay — Appears to have shut outside AI agents, including Meta's Muse, out of its listings — the same weekend a user's Muse-built daily watch digest stopped pulling eBay data — while Amazon blocks agents outright and Shopify invites them in. Agent access to marketplaces is becoming a strategic decision, not a technical one; if you sell through these channels, who may shop on your customers' behalf is being decided now. eBay confirmed it has restricted agent access in certain cases, including scraping; for scale, DoorDash delivered 993 million orders last quarter, of which only a few hundred came from agents. The Verge
Risk the board will raise
Vendor breach exposed Notion AI chats. An intrusion at Fireworks AI, a behind-the-scenes supplier to Notion, exposed chat transcripts from Notion's AI assistant for fewer than 1% of workspaces; the transcripts carried no names or workspace labels. Staff paste sensitive things into AI chats, and those chats sit with vendors you may never have heard of — ask the same question of your own AI toolchain before the board does. Fireworks flagged possible access on November 6 and confirmed the dataset on November 8; Notion has cut Fireworks off and deleted the data it held. Notion Help Center
“We recommend that you treat this data as if it were exposed.”
Hedge fund blames AI tools for a code theft. Point72's quantitative arm Cubist alleges in a lawsuit that a fired lead research analyst used Claude and OpenAI's Codex to copy proprietary trading code before leaving for a competitor — an allegation, not a finding. The novel board risk is not agents attacking you, but staff using everyday AI tools as the quickest way to walk out with your code; check what your offboarding process actually covers. The filing, reported by Bloomberg, names tools from both Anthropic and OpenAI as the route. Techmeme / Bloomberg
The big AI stories today
Grok Build — Grok Build's shareable links exposed confidential data from thousands of apps, including 24 built inside SpaceXSI itself. Ask where your staff's AI-built internal tools live, and who else can open them. Ars Technica
Claude Fable 5.2 — Anthropic released two new models priced about 20% below OpenAI's GPT-6 Astra, days before its listing. The price war among the big vendors is now explicit — a good moment to revisit your model contracts. Fortune
Lenny's Newsletter — An analysis of late-2026 tech hiring finds product managers gaining most, engineer demand plateauing and designers struggling. Worth a read before you approve your next headcount plan. Techmeme / Lenny's Newsletter
One takeaway
Strands Agents blog · yesterday
The savings are inside; the customers are not
Yesterday split the agent story in two: outside companies' walls, customer-facing agent demand remains a promise, while inside them the savings are real, specific and already banked. Ask your team which high-volume internal task should move to a cheaper, narrower model next — and treat any plan that counts on agents arriving as customers as unproven.