Signal SentrySearchUpdated Nov 14, 05:30 UTCAM Drop
Drop #085 · Sat, Nov 14, 2026 · Morning

Agents need a second checker

10-second version
  1. Google and Momentic both put a separate checker on agent work; Momentic says 9.6% of 1,000 AI-written tests were useful.
  2. CISA flags a CVSS 9.9 prompt injection flaw in Jet Admin that can leak .env credentials; n8n lists an OAuth hijack fix.
  3. Investors say Nvidia's startup checks may shrink after Huang; a House chair says Huang agreed to testify within weeks.
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. New this week: teams split the agent that builds from the one that checks

    If you assumed
    that a coding agent whose own tests pass has finished the job
    Now test
    whether a checker that never saw the builder's context, working from the original requirements, finds bugs your agent's own tests missed
    This week
    Take one agent-built feature from this sprint and have a separate agent or a person test it live against the spec

    Based on:Momentic: only 9.6% of 1,000 AI-written unit tests were useful checksPreprint: front-end agents scored 87.69% on tasks but 51.52% on whole appsGoogle says its /teamwork agents formalized the Weil conjectures, opens tools to users3 primary sources

  2. Agent spend is being measured per solved task and capped per agent

    If you assumed
    that price per token tells you what an agent will cost
    Now test
    what each coding-agent task that passed its tests actually cost, and whether you can cap spend for each agent or user
    This week
    Pull last week's agent runs and divide total spend by the runs whose tests passed

    Based on:AWS shows NeMo Switchyard routing coding-agent calls between two models on BedrockGoogle adds agent telemetry, token quotas and evals across BigQuery and Looker2 primary sources

443models priced on OpenRouter
+124★stars today on the top repo
4,166open roles at AI labs
14 ddays to the next AI-law deadline
Ranked by impact

Top 5 that matter

1
Agents

Google says its /teamwork agents formalized the Weil conjectures, opens tools to users

Google Research says /teamwork agents on Gemini 4 models produced a machine-checked formalization of the Weil conjectures: about 1.8 million Lean lines over 28,800 subagent calls. Patterns like /boost are in Antigravity now.

Why it matters: Google's design hands each agent only the files it needs and has critics recheck work from clean contexts. Try that split before scaling one long-running agent.

On the Frontier watch →
2
Agents

Momentic: only 9.6% of 1,000 AI-written unit tests were useful checks

In a Stack Overflow post, Momentic says a human-reviewed audit of 1,000 AI-written tests in its monorepo found 9.6% useful. Its separate QA agent found 3-5 times as many bugs as Codex or Claude Code alone, on four open-source apps.

Why it matters: Momentic sells this kind of verifier, so weigh its benchmark with that in mind. The check for you: does anything other than the coding agent test what it built?

3
Security

CISA: prompt injection in Jet Admin Cloud Platform can leak .env credentials

CISA published advisory VA-26-317-01: a prompt injection flaw (CVE-2026-123965, CVSS 9.9) in all versions of Jet Admin Cloud Platform lets anyone holding a free account remotely expose .env files and steal credentials.

Why it matters: An AI feature that can read server config turns one prompt into a credential leak. If you run internal-tool builders, check which files their AI can reach.

Confirmed Way in: a prompt injection from a free account that exposed .env configuration files

Controls that would have helped:

4
Business

VCs tell TechCrunch Huang's exit leaves a hole in AI startup funding

Nvidia joined 105 funding rounds this year, often as the largest investor, TechCrunch reports citing PitchBook. Investors expect interim CEO Colette Kress to be more conservative, with smaller checks and fewer new bets.

Why it matters: If you rely on a neocloud or open-weight lab that Nvidia backs, ask about its runway without new Nvidia money; investors expect valuations to drop.

5
Agents

Claude Meetings joins Google Meet calls in beta and skips calls with outside guests

Anthropic says Claude can join video calls as a visible participant, answer questions, take notes and build docs. The beta is rolling out to Team and Enterprise plans, on Google Meet first.

Why it matters: By default it stays out of calls with outside guests and saves transcripts, not audio. Decide who may turn on auto-join, and confirm attendees accept transcription.

Models

New in models

See all Models →
Models

Perplexity releases Decider v1.3 decision models under an Apache license

Why it matters: Jobs like routing, judging and screening tool output for prompt injection may not need a generative model. Price one of these against the LLM calls you use today.

More detail: Perplexity releases Decider v1.3 decision models under an Apache license

Perplexity says Decider v1.3 and Decider Flash v1.3 return probabilities for yes/no, multiple-choice and score questions in one pass, and add locate for pulling out text spans. Both are on Hugging Face, OpenRouter and its API.

Agents

Agents at work

See all Agents →
Agents

Preprint: front-end agents scored 87.69% on tasks but 51.52% on whole apps

Why it matters: A high per-task score can hide apps whose core workflows fail. Judge coding agents on whether every must-have flow works in a single run.

More detail: Preprint: front-end agents scored 87.69% on tasks but 51.52% on whole apps

Researchers report in a preprint that on JoyFrontBench, five coding models under OpenHands averaged an 87.69% task score but a 51.52% application success rate. Model rankings shifted between the OpenHands and OpenCode harnesses.

Agents

Preprint: auto-tuning a multi-agent harness lifted accuracy, left orchestration failures

Why it matters: Automated harness tuning may patch symptoms without fixing how agents delegate and share information. Review the handoffs directly rather than trusting a higher score.

More detail: Preprint: auto-tuning a multi-agent harness lifted accuracy, left orchestration failures

Researchers report in a preprint that MULTILOOP, a loop where a coding agent edits a multi-agent harness, lifted DeepSearchQA accuracy 4.6% over the base harness. Orchestration failures largely persisted.

On the Frontier watch →
Agents

Claude Code Projects' coordinator now posts status updates for parallel workers

Why it matters: Every parallel worker draws on the same plan allowance, and several high-effort threads can drain it fast. Set worker effort levels before you fan out work.

More detail: Claude Code Projects' coordinator now posts status updates for parallel workers

AlphaSignal reports Anthropic's Claude Code Projects coordinator now answers questions and reports thread progress in chat, and adds read-only thread links, reminders, /btw and /fork. Accounts are capped at 200 new threads a day.

Agents

Google adds agent telemetry, token quotas and evals across BigQuery and Looker

Why it matters: By design, logging skips turns whose sources use row-level security or masked data. Check that gap against your audit needs before relying on these logs.

More detail: Google adds agent telemetry, token quotas and evals across BigQuery and Looker

Google Cloud says conversation logging to customer-owned BigQuery tables and agent monitoring dashboards are GA, with daily token quotas per project, user or agent in preview. ADK and LangGraph agents can stream traces too.

Agents

AWS shows NeMo Switchyard routing coding-agent calls between two models on Bedrock

Why it matters: AWS argues routing should be judged by cost per solved task, not price per token. Measure that on your own agent runs before switching models.

More detail: AWS shows NeMo Switchyard routing coding-agent calls between two models on Bedrock

AWS's sample runs LiteLLM with NVIDIA's open-source Switchyard plugin, which reads recent tool results to pick Claude Opus 5.5 or Nemotron 3 Super for each request. AWS labels it sample code for non-production use.

Security

Attacks and flaws

See all Security →
Security

Canada's Cyber Centre flags an OAuth account hijack flaw in n8n

Why it matters: n8n workflows often hold OAuth tokens for many services. If you self-host it, check your version against the fixed releases.

More detail: Canada's Cyber Centre flags an OAuth account hijack flaw in n8n

The Canadian Centre for Cyber Security's advisory AV26-1160 lists an OAuth account hijack via an unauthenticated callback in n8n's end-user credential connect flow, affecting versions before 2.45.5, 2.46.1 and 3.0.1.

Business

Money and deals

See all Business →
Business

Amazon drafted a plan to index products sold across the web, Reuters reports

Why it matters: If shopping agents lean on one retailer's index, that retailer shapes what they see. Online sellers should watch whether Amazon starts crawling their product pages.

More detail: Amazon drafted a plan to index products sold across the web, Reuters reports

The Verge, citing Reuters, says an internal Amazon document describes a World Product Graph including goods Amazon doesn't sell, partly so AI agents use its index. Amazon says such brainstorming documents may not reflect its plans.

Business

Cohere completes its AISLE purchase and brings in cubic's code review team

Why it matters: Model vendors are bundling security and code review for on-premises and air-gapped buyers. If you need those, ask Cohere what runs inside your own environment.

More detail: Cohere completes its AISLE purchase and brings in cubic's code review team

Cohere says it closed its acquisition of AISLE, an autonomous vulnerability-fixing company, and is welcoming the team behind cubic's AI code review platform. It counts four deals this year, after Aleph Alpha and Reliant AI.

Business

OpenAI folds its stand-alone robotics team into research and Applied, Fortune reports

Why it matters: OpenAI is betting general models can run robots without purpose-built brains. Robotics buyers should ask vendors what on-site data their robots gather that a general model lacks.

More detail: OpenAI folds its stand-alone robotics team into research and Applied, Fortune reports

OpenAI confirmed to Fortune it moved robotics researchers under Chief Research Officer Mark Chen and its applied robotics and hardware teams into Applied. OpenAI says models like GPT-6 Astra apply directly to robotics.

Laws

Law watch

See all Laws →
LawsFollowing since Nov 12, 2026in 5 Drops

Rep. Mast says Huang agreed to appear before House committees within weeks

Why it matters: Scrutiny is widening from Nvidia to Commerce's export enforcement. If your compute plans depend on chips bound for Asia, expect sharper questions on end users.

More detail: Rep. Mast says Huang agreed to appear before House committees within weeks

CNBC reports House Foreign Affairs chair Brian Mast said Huang agreed to testify, likely under oath and before Thanksgiving. Sens. Warren and Coons asked officials to preserve records on Nvidia chip exports to China.

Laws

FAA expands its $875 million SMART AI traffic advisor to all 22 US centers

Why it matters: A federal template for AI advice at scale: people validate before anything reaches operators. Compare your own high-stakes AI rollouts against that review step.

More detail: FAA expands its $875 million SMART AI traffic advisor to all 22 US centers

Ars Technica reports the FAA says SMART rerouted 180 flights over Washington, saving about 10 minutes each on average. A DC team validates its advice, controllers may decline it, and their union says it wasn't consulted.

Today's Talk

What people are arguing about

See all Today's Talk →
Today's Talk

HN weighs a No AI Slop manifesto and whether AI will speed up AI development

Why it matters: Builders are split on the quality of AI output versus its pace. Worth a skim if your team is writing rules for AI-generated content.

More detail: HN weighs a No AI Slop manifesto and whether AI will speed up AI development

On Hacker News, No AI Slop: A Manifesto drew 189 points and 108 comments, and a thread titled The avalanche that hit math is coming for AI development drew 101 comments.

Play the NewsShip It or Block ItAbout 2 minutes, built from this Drop's items.