AI selloff day two; OpenAI trims its round; DoorDash counts its agent orders

DoorDash's agent orders — The delivery company told investors that outside AI agents — services that order on a customer's behalf — have produced only a few hundred orders in total, out of nearly a billion orders last quarter. CEO Tony Xu said he will not fight the technology but will not be delusional about it either. If you are budgeting for agentic sales channels, demand a usage number first. This is the first hard number from a named company on how many customers actually buy through agents, and it is close to zero. Amazon blocks personal agents while Walmart welcomes them — the market has not decided, and neither should your plans. DoorDash's own in-app assistant, by contrast, builds baskets five times faster and lifts basket value by nearly 50%. TechCrunch

“if you look at where AI agents are today, we're seeing maybe several hundred orders in total from some of these third-party agents”

Bank of America and Capital One — Two of the biggest US banks told an AI conference that they contain AI costs by sending simple work to small, cheap models rather than the most expensive ones. Bank of America is doubling its AI budget next year regardless, and Capital One says smaller models cut its response delays fivefold and lifted dealership engagement by 50%. Cost control and bigger AI budgets can go together — the lever is routing, not retreat. Ask your engineering lead whether simple internal tasks are being routed away from the priciest models. Bank of America's routing layer, Orchestra, is a small classifier model that sends each query to the cheapest model that can handle it. CIO Dive

StackOverflow — A widely shared thread notes that StackOverflow, once the default place programmers asked questions, now receives fewer than 1,000 questions a month — fewer than it averaged every month in 2024. Commenters credit AI tools that answer directly. The quiet loss for companies is fresh, written-down engineering knowledge. This is what the shift to AI answers looks like measured: when people stop asking colleagues in public, know-how stops being captured. Worth checking where your team's questions now go unanswered. One commenter noted the platform's total questions last year were fewer than the 2024 monthly average. data.stackexchange.com · from the Hacker News discussion

OpenAI's funding round — OpenAI has cut its target from about $30 billion at a $1.4 trillion valuation to roughly $20 billion at $1.25 trillion, according to unconfirmed press reports, and has about $11 billion committed so far, with more expected in coming weeks. Even the AI market leader is finding money scarcer this month. A useful signal when you negotiate long contracts with any AI vendor: their own cost of capital just went up, and pricing behaviour may follow. Both reports attribute the change to investors pulling back commitments during the AI selloff. Financial Times via Techmeme

Anthropic's credit rating — Blackstone, anchor investor in Anthropic's $63 billion chip-financing deal, has told backers the company may emerge from its coming listing with a 'high' junk credit rating rather than an investment-grade one, reversing earlier guidance. The Hacker News thread's view is blunt: no surprise, since the labs spend far more than they earn. It signals Anthropic will lean harder on partners and debt markets to fund its build-out, and it is one more wobble to an IPO some expected to be the largest listing ever. Anthropic filed to go public on Monday and could list next month. ft.com · from the Hacker News discussion

AWS — Amazon's cloud arm grew 39% year over year last quarter, and CEO Andy Jassy told investors it could reach $1 trillion in revenue within five to ten years. Its backlog of signed contracts grew to $578 billion, including new agreements with Meta, Nvidia, OpenAI, Salesforce and Oracle. The demand side of AI is real even as investors punish the spending — the sensible read is that the boom is being financed badly, not imagined. Amazon says it is installing an AI accelerator roughly every 14 seconds. CIO Dive

Reddit — Reddit said it will become a mostly logged-in-only service, blaming search referrals that fell by over 1 million per week in October as Google's AI answers keep people from ever clicking through to websites. Revenue still rose 53% to $895 million, and 80% of its ad money already comes from logged-in users. The first big consumer platform to restructure itself around the loss of search traffic. If search referrals matter to your sales pipeline, model this risk now rather than later. CEO Steve Huffman said Google's AI Overviews 'are not traffic drivers'. Business Insider

Claude's security research — Anthropic disclosed that an internal research version of Claude found and chained together serious flaws in Chrome, Safari, Android, Windows and Linux; 62 assigned Chrome bugs were rated critical or high and are already patched. Specialist attack skills once held by a handful of security teams are now cheap and widely available — to attackers as much as defenders. Ask your security lead what changes when the attacker's cost per exploit becomes a rounding error. All vendors were given the standard 90 days to patch before details were published. Anthropic

Ghost student fraud — The US Justice Department charged five people in its first broad crackdown on 'ghost students' — fraudsters who use stolen identities and AI to pose as college students and drain federal loans. Officials identified over 830,000 suspect aid accounts and seized more than $750,000 in cryptocurrency. AI-enabled identity fraud at scale is now a live enforcement area. If your business verifies customers online, expect regulators to demand harder checks — and expect some real customers to be caught in them. Officials put confirmed losses tied to the charged individuals at more than $32 million, with millions more suspected. Ars Technica

AI scraping opt-out — OpenAI, Google, Microsoft and Anthropic will honour one shared website setting that blocks their AI crawlers at once — useful if your content is being scraped. The Verge

Spotify AI remixes — Spotify reportedly plans a roughly $5-a-month AI remix tool for early 2027, unconfirmed — a test of whether consumers will pay extra for AI features. Financial Times via Techmeme

Ask HN: AI fatigue — A Hacker News thread on AI-era fatigue drew 132 comments describing engineer burnout from managing agents — worth watching your own team's load. Hacker News