LLM Digest
Subscribe

AI Daily Recap

23 articles · 6 categories

View as JSON
‹

The finishable daily brief

What happened in AI — Nov 19, 2026

Thursday, Nov 19, 2026
23 articles · 6 categories

read top to bottom · then stop

In 30 seconds

  • AgentDojo scored 433 runs as no attack even when the agent paid the attacker; re-check security claims built on it.
  • Gemini API Tool Retrieval (preview): set defer_loading and add tool_search to load tool schemas only when needed.
  • An autonomous Codex campaign reported 507 reproduced Webmin vulnerabilities for about $4,578 in tokens.
  • Vercel's eve agent framework reached 1.0; LangSmith Tool Gateway entered public beta for governed tool calls.
  • agent-framework follow-up: PyPI quarantined Microsoft's publisher; the attacker may still have publish access.
  • DeepSeek dropped separate fees for model-triggered search and web fetch.

The tool layer moved out of the prompt today. Gemini now loads tool schemas on demand, LangSmith put credentials and permissions behind a gateway, and Vercel shipped eve 1.0.

Security evidence cut both ways. An audit found AgentDojo scored 433 real attacks as misses, while an autonomous Codex run reported 507 Webmin bugs for about $4,578.

Prefer to watch?

LLM Digest Daily Explainer: Nov 19, 2026

How today's stories work, in 2:00. Loads from YouTube when you press play.

Tool layers: load less, govern more 4 items

Vendors moved tool handling out of the prompt: schemas load on demand, credentials sit behind a gateway, and output JSON gets compressed in a proxy.

Agent harnesses grow up: frameworks, state, sandboxes 5 items

Agent runtimes are becoming infrastructure with GA labels, in-house standards, and durable state, not per-team experiments.

Eve 1.0 – Agent framework from Vercel

Hacker News AIDetails

Vercel's open-source agent framework hits GA after 230+ preview releases, adding approval controls for actions and an eval suite for how well coding agents build with it.

Aviator: HubSpot's Agent Harness

Hacker News AI

HubSpot explains why it built one internal agent harness in 2025 after teams kept making divergent model and architecture choices.

Docker AI Rebranded

Hacker News AI

Docker docs now lead with agent sandboxes: scheduled tasks run in a fresh cloud sandbox that is deleted when the task ends or times out.

Evals that check what the agent did 4 items

Today's eval work audits actions and traces instead of trusting a model's reply text or one automated verdict.

Security: AI finds bugs, supply chain still leaks 3 items

Autonomous auditing got cheap enough to run on a whole codebase while package and model supply chains stayed the soft spot.

Agents doing production engineering work 3 items

Large operators described agents that run inside their own engineering loops, from performance fixes to RL rollouts.

Token prices and compute deals 4 items

Chinese and Indian labs competed on price while compute commitments shifted in both directions.

You are caught up for this edition