2026-11-05
By Vadym · Generated with AI, curated by me
Today’s theme is hype meeting its own usage data. Uber admitted almost nobody books rides through AI agents, and Apple is quietly rationing the AI tools it told staff to lean on all year. Meanwhile Anthropic is building infrastructure nobody’s required to adopt, OpenAI is lobbying two years ahead of an election, Meta is buying its way into agentic coding instead of out-building it, and Nothing is betting its first AI device doesn’t need a screen at all.
Meta confirmed it bought Conductor, a 2024-founded startup whose product deploys multiple AI agents into parallel workspaces so engineers can work on different parts of a codebase at once — using models from labs like Anthropic and OpenAI rather than its own. Core features keep running through early next year, but Conductor’s cloud service shuts down November 20, after which the company starts deleting user cloud data. Founder Charlie Holtz says something new is coming for customers by early next year.
OpenAI’s chief global affairs officer Chris Lehane published an essay arguing the 2028 presidential race will be decisive for AI policy, comparing the moment to the New Deal. His five asks: worker tax incentives and profit-sharing, child-safety guardrails for AI companions, enforceable safety standards with a shutoff capability, and support for data center buildout while opposing moratoriums. AI Now Institute’s Kate Brennan called the framing “disingenuous,” citing OpenAI’s past opposition to California’s SB 1047.
Nothing CEO Carl Pei posted that the company’s first “AI-native product” ships in December. He said it has no screen, isn’t an app, and is pitched as a productivity tool — distinct from the AI wearables like glasses and pins the rest of the industry is chasing. No pricing or feature details yet.
On Uber’s Q3 earnings call, CFO Balaji Krishnamurthy said only four rides were booked via an AI agent during the week of Halloween — its busiest week for agent activity — against roughly 320 million total trips and orders that same week. Uber Eats saw about 500 agent orders in the same window. CEO Dara Khosrowshahi called it “very, very early and very limited traction,” noting that booking a ride in-app already takes about 50 seconds.
Bloomberg reports Apple cut engineering teams’ daily AI spending cap from $300 to $200 in October, with further cuts planned to as low as $100 a day for engineering and $25 a day outside it. Apple hasn’t commented on the specifics or the reasoning.
Anthropic released FlashProve, an open-source system for cryptographically verifying which model produced a given output — so-called “provable inference.” Labs publish a fingerprint for each model; outputs can be matched to it without revealing model weights, and verification is lightweight enough to run on a local device. Anthropic published a spec and working code on GitHub but has no current plans to deploy it itself or anywhere else.
Step back and today splits into two buckets: hype deflating in daylight, and groundwork for the next phase getting laid quietly underneath it. Uber’s own numbers and Apple’s own budget cuts are both companies admitting, in their own data, that the agent-and-productivity story hasn’t caught up with the spend yet. Meanwhile Anthropic is building trust infrastructure nobody’s required to use, OpenAI is laying political groundwork two years before anyone votes, and Meta and Nothing are both betting on form factors — acquired agents, a screenless gadget — rather than waiting for the obvious ones to pay off. The honesty this week is coming from the usage data, not the keynotes.
— Boba
Curated by Vadym