Signal SentrySearchUpdated Nov 12, 05:30 UTCAM Drop
Drop #081 · Thu, Nov 12, 2026 · Morning

United shuts out AI agents; Google prices fast decisions

10-second version
  1. United bars unauthorized AI agents from booking its flights, including through agencies and aggregators.
  2. Google's Gemini Decisions API scores up to 1,000 questions in one pass; AWS maps Codex onto Bedrock controls.
  3. Nvidia's board weighed ousting Jensen Huang, CNBC confirms; Anthropic and a preprint tie agent memory to its source runs.
Our read

What changed

Every signal →

Our reading of this Drop's sources: what to rethink, and one thing worth checking this week. Not a prediction.

  1. Agent blocks now reach partner channels, not just a merchant's own site

    If you assumed
    that an agent shut out of a merchant's site can still finish the purchase through an agency or aggregator
    Now test
    whether each merchant your agent buys from bars unauthorized agents across its resellers too, as United says it will
    This week
    List the merchants your agent transacts with and check each one's current terms on third-party agents

    Based on:United bars third-party AI agents from booking its flights

  2. Agent memory is only as good as the record behind it

    If you assumed
    that notes an agent saves from past runs are safe to reuse as written
    Now test
    whether your agent can trace each saved lesson back to the run it came from, and correct it
    This week
    Keep original transcripts next to agent memory notes, and spot-check a few notes against them

    Based on:Anthropic: stronger Claude models pick what to remember; dreaming corrects bad notesPreprint: agent memory works better when each lesson links back to its source run2 primary sources

The Drop in 60

This Drop in 49 seconds

Tap to play. It never plays by itself, and the captions are in the picture.

Transcript

United just shut out AI agents. United says unauthorized AI agents can no longer book its flights, even through aggregators. Google's new Decisions API scores up to 1,000 questions in one pass. AWS showed how to run Codex through Bedrock with a least-privilege pilot. CNBC confirmed Nvidia's board met on whether to oust Jensen Huang, after reports of two affairs. And LangChain says deals with an agent-built document advanced 47 percent of the time, versus 16 percent without. If your agent buys for users, check every merchant's rules on third-party agents before it tries to pay.

Made with AI from this Drop's sourced items. The facts come from the sources linked on each item. Script: Claude · Voice: OpenAI gpt-4o-mini-tts · Visuals: motion graphics

439models priced on OpenRouter
+173★stars today on the top repo
4,172open roles at AI labs
16 ddays to the next AI-law deadline
Ranked by impact

Top 5 that matter

1
Agents

United bars third-party AI agents from booking its flights

United says unauthorized personal AI agents may no longer book or manage trips, on its own platforms or via agencies and aggregators, CBS News reports. Agents can still show United flights, The Verge says.

Why it matters: If your agent buys on a user's behalf, check each merchant's terms: United cites exposed credentials, unintended purchases and federal check-in rules as its reasons.

2
Models

Google's Gemini Decisions API scores up to 1,000 questions in one pass

Google says Gemini Decision 1 scores fixed options instead of generating text, returning probabilities in 150-250 ms for $0.10 per 1M input tokens, with output tokens free. Inputs can mix text, images, audio, video and PDFs.

Why it matters: A cheap, fast check for routing, tool choice and policy gates. Each answer carries a confidence score, so test where your threshold should hand a case to a person.

3
Agents

AWS shows how to run OpenAI Codex through Amazon Bedrock under company IAM controls

AWS says Codex can send inference to Bedrock for supported OpenAI models while still running on the developer's machine. It advises a small least-privilege pilot, a two-user audit separation check and a tested rollback.

Why it matters: Teams on AWS can bring Codex under existing sign-in, logging and billing. AWS also says to keep the provider setting where a repository can't change it.

4
BusinessFollowing since Nov 9, 2026in 4 Drops

Nvidia's board meets on whether to oust Jensen Huang, CNBC confirms

CNBC confirmed Nvidia's board met Wednesday on Huang's future after reports of two affairs, one with an alleged Chinese spy. Bloomberg reports an outside firm is seeking candidates, TechCrunch says; there is no obvious successor.

Why it matters: Nvidia's leadership shapes chip supply and export talks. If your capacity plans rest on Nvidia, watch who leads and what the board's Wachtell Lipton probe finds.

5
Agents

LangChain's sales agent made 1,540 deal documents in 12 weeks, the company says

LangChain says deals where reps sent an agent-built document advanced 47% of the time versus 16% for matched deals without one. A rule to stop self-review after one clean pass cut a one-pager run from 18 to 7 minutes.

Why it matters: It is LangChain's own sales team, and matched deals are not a controlled test. The lesson to check: give review loops a stopping rule, and put effort into context, not prompts.

Agents

Agents at work

See all Agents →
Agents

Anthropic: stronger Claude models pick what to remember; dreaming corrects bad notes

Why it matters: Anthropic found fixed memory rules can hurt strong models. Check whether your agents' saved notes are ever tested against the record they came from.

More detail: Anthropic: stronger Claude models pick what to remember; dreaming corrects bad notes

In Anthropic's tests, Claude Fable 5 kept nine memory files and earned four Pokémon badges, while Sonnet 3.7 wrote 413 files and earned one. Rereading a session transcript let Claude catch and rewrite a wrong note.

On the Frontier watch →
Agents

Preprint: agent memory works better when each lesson links back to its source run

Why it matters: Lessons stored without evidence can mislead an agent. If you distill runs into notes, keep a link to the original trace so the agent can check before acting.

More detail: Preprint: agent memory works better when each lesson links back to its source run

Researchers report in a preprint that agents can fail tasks they would solve without memory by following wrong or misread lessons. Their WikiTrace ties each lesson to its source run and beat other memory methods on five benchmarks.

On the Frontier watch →
Agents

MCP moves to two spec releases a year; next version targeted for March 30, 2027

Why it matters: Teams building on MCP can plan upgrades around fixed dates. Mark the 2027-03-30 version and test SDK betas against your servers when they appear.

More detail: MCP moves to two spec releases a year; next version targeted for March 30, 2027

The MCP maintainers say spec releases will generally land on the last Tuesday of March and September, with soft and hard freezes and SDK betas before each. Proposals that are not ready move to a later release.

On the Frontier watch →
Agents

AWS releases an official local emulator for its services, written by a swarm of agents

Why it matters: Cheaper offline tests for AWS code, and a large public example of agent-written infrastructure. Check its coverage of the services you use before swapping out mocks.

More detail: AWS releases an official local emulator for its services, written by a swarm of agents

Simon Willison says AWS's new aws-emulator lets tests run against a local copy of many AWS services without a paid key. It is written in Rust by agents, with 29,684 commits in its first three weeks, and needs Docker on macOS or Windows.

Security

Attacks and flaws

See all Security →
Security

Hackers use AntiFox AI agents in AWS credential-theft attacks, BleepingComputer reports

Why it matters: Cloud keys remain a prime target. Check which AWS credentials sit on laptops and inside agents, and how long each stays valid.

More detail: Hackers use AntiFox AI agents in AWS credential-theft attacks, BleepingComputer reports

BleepingComputer's headline says attackers are using AntiFox AI agents to steal AWS credentials. We could not read the full report, so how the attackers got in is not described here.

Business

Money and deals

See all Business →
Business

OpenAI's quarterly revenue tops $10 billion for the first time, Bloomberg reports

Why it matters: TechCrunch ties the gain to tighter pricing, enterprise products and agents. Compare OpenAI and Anthropic run rates with care: they count partner sales differently.

More detail: OpenAI's quarterly revenue tops $10 billion for the first time, Bloomberg reports

Bloomberg, cited by TechCrunch, says demand for OpenAI's recent models lifted Q3 sales past $10 billion from $6.7 billion in Q2, while adjusted operating margin improved from -150% to -109%.

Business

TSMC board approves up to $1 billion in stakes in OpenAI and Anthropic, Reuters says

Why it matters: The firm that makes most of Nvidia's AI chips now backs two of the biggest end buyers. Weigh circular deals like this when reading AI demand signals.

More detail: TSMC board approves up to $1 billion in stakes in OpenAI and Anthropic, Reuters says

Per Reuters, cited by Ars Technica, TSMC plans up to $500 million in OpenAI preferred stock and $500 million in Anthropic shares at its expected IPO price. Analysts warn such supplier-customer deals blur real demand.

BusinessFollowing since Nov 11, 2026in 3 Drops

CoreWeave leans on about $9 billion in prepayments and pricier short-term deals

Why it matters: Compute buyers are being asked to pay up front or accept shorter, pricier terms. If you rent GPUs, check renewal pricing and prepayment terms before signing.

More detail: CoreWeave leans on about $9 billion in prepayments and pricier short-term deals

TechCrunch says CoreWeave expects about $9 billion in prepayments and deposits over three months. Its CEO says short-term capacity goes for over $40 million per megawatt yearly, vs $20 million to $25 million on longer contracts.

Laws

Law watch

See all Laws →
Laws

Lawmakers press for Huang testimony and tighter chip export bills, CNBC reports

Why it matters: No bill has passed. Anyone selling or buying advanced chips abroad has three to watch in defense bill talks: the AI Overwatch, MATCH and Chip Security Acts.

More detail: Lawmakers press for Huang testimony and tighter chip export bills, CNBC reports

CNBC reports House committee chairs asked Huang to testify on Nov. 18 and say subpoenas are possible, while Sens. Warren and Banks urged adding the AI Overwatch Act to the defense bill to curb chip flows to China.

Repos

Climbing fastest

See all Repos →
ReposFollowing since Nov 11, 2026in 2 Drops

lot-map shows Claude Code's tool calls and subagents live beside the chat; 427 stars

Why it matters: A quick way to see what a coding agent does in a session, useful before granting it more access.

More detail: lot-map shows Claude Code's tool calls and subagents live beside the chat; 427 stars

The MIT-licensed lot-map, created Nov. 9, gained 173 stars since our last check. It lights up Claude Code's tool calls, model steps, hooks, skills and subagents as they happen, runs locally and comes in five languages.

Repos

onlyread makes AI browser agents read-only by refusing writes; 86 stars

Why it matters: A simple guard for trying browser agents on real accounts: let them look, but not act, until you trust them.

More detail: onlyread makes AI browser agents read-only by refusing writes; 86 stars

onlyread is a Chrome extension that blocks sends, posts, edits and deletes before they leave the browser. Licensed AGPL-3.0 and created Nov. 5, it gained 31 stars since our last check.

Today's Talk

What people are arguing about

See all Today's Talk →
Today's Talk

HN asks why AI coding has not made teams ship faster; 102 comments

Why it matters: If you measure AI by code written, this is a prompt to check what actually reaches production.

More detail: HN asks why AI coding has not made teams ship faster; 102 comments

A Hacker News thread (109 points) discusses Christian Bird's Oct. 30 UC Irvine talk, Why aren't we shipping faster now?, which draws on studies including an NBER paper on writing versus shipping code.

Play the NewsAI ConnectionsAbout 3 minutes, built from this Drop's items.