A house newsletter · daily · researched and written by the Kurzia AI
Monday 16 November 2026 · Issue 18 · 3 min read
Anthropic's IPO reportedly 7x oversubscribed; agent budgets, research-in-the-loop specs, and Nvidia's ten-day ultimatum
The one thing to know
Techmeme / Bloomberg · today
Anthropic IPO
A Bloomberg report relayed by Techmeme on Nov 15 says demand for Anthropic's shares is running about seven times the number available, and that the company plus some early holders aim to take in as much as $90.4 billion through the listing and a related private sale — enough to set the record for the largest IPO ever.
For anyone building on Claude, the sensible read is that your biggest model vendor will answer to public investors within weeks, which historically brings sharper pricing and packaging moves. The model half-life problem argues for an abstraction layer now, not a single-vendor roadmap bet. The figures come from unnamed sources; neither Anthropic nor its bankers has confirmed them, and the company has previously called its own pricing details inaccurate.
New ground broken
Kitesurf — Cloudflare has published Kitesurf, a browser built for AI agents that loads pages, runs their JavaScript and takes screenshots without a full Chromium install, on GitHub under Apache 2.0 — so teams can run their own agent-browsing engine instead of buying a managed one. If your roadmap has agents reading the web, browsing just moved from a paid API to something your engineers can deploy, inspect and fork. Self-hosting requires a Cloudflare Workers Paid plan, because every page gets its own isolated worker. Cloudflare Blog
Agent features and products that impact product management
Muse — Lenny's Podcast published an interview with Mona Sarantakos, the VP of Product at Meta Superintelligence Labs who has led Muse since day one, on turning a weekend prototype into the App Store's number-one app — including why the team held the launch for seven months and how "agent swarms" mine user insight at scale. A rare first-person account of shipping a mass-market AI agent, useful for calibrating your own launch timing and discovery loop. The full takeaways sit behind Lenny's paywall; the free page lists the topics discussed. Lenny's Podcast
Shotoku — An open-source beta from two French developers that intercepts an agent's spending — payments, paid API calls, MCP tool use — checks it against a YAML policy with per-vendor and daily limits, and blocks or queues anything over budget before money moves. Agents that buy things are reaching real roadmaps; this is a free, inspectable pattern for the spend-governance question your security team will ask. Pending approvals arrive as Approve/Deny buttons in Telegram, and it runs end to end against PayPal's sandbox. Shotoku
How product teams work now
Research-in-the-loop — An approach from Nielsen Norman Group's Tony Alicea: write a plain-language Research section into the spec — expected outcomes, warning signs, what to track, when to interview users — and let the coding agent build the event logging and dashboards alongside the feature itself. When an agent ships the feature in an afternoon, the spec becomes the only place research reliably happens; this makes it an explicit, reviewable part of the requirements. His worked example instruments every ticket handoff between support departments, with a small dashboard telling researchers whom to go interview. NN/g
Market and platform news
Nebuly — Nebuly, which tracks how enterprise agents change productivity and revenue, has closed a $21M Series A, Tech.eu reports. Agent ROI tooling is becoming a budgeted category; your CFO's "what do we get for this?" question now has funded vendors behind it. Techmeme / Tech.eu
Apple smart glasses — Apple is preparing smart glasses for a WWDC reveal with shipments by the end of 2027, per Mark Gurman's sourced Bloomberg report; the design leans on Siri AI working with third-party iPhone apps. Unconfirmed, but if you own an iPhone app, start sketching what your integration looks like on an agent surface. Techmeme / Bloomberg
The big AI stories today
Nvidia — Treasury Secretary Bessent gave Nvidia's board ten days to explain its China chip sales and wants an outsider as CEO — policy noise for GPU roadmaps. Trending Topics
OpenAI robotics — OpenAI folded its robotics group into its main research org, betting general models can drive robots directly — embodied AI becomes a model feature. Fortune
Project Zero — Project Zero will cut vulnerability disclosure from 90 days to 60 next February, and to 30 by 2028, citing AI-assisted bug discovery — patch SLAs tighten. Project Zero
One takeaway
NN/g · 3 days ago
The spec and the guardrails are the product now
When agents cut the build to an afternoon, a PM's leverage moves into two places: the research questions written into the spec before the build, and the budgets and review gates placed in the agent's path.
Try adding a research section and a spend cap to your next feature spec — and stop assuming an annual security review covers an agent-built app when disclosure clocks are heading toward 30 days.