Simon Willison highlights an essay by Jeremy Scheff on what coding agents take away from programmers who loved solving logic puzzles. Willison notes that developers who mainly care about building working systems see the same tools very differently.
Nace AI has released Drex Max, a decision model that reads text and screenshots and returns calibrated probabilities for computer-use agents. The company reports a score of 67.59 on Decision Index 0.3.2, at $0.06 per million input tokens.
Town, a startup whose AI agents collaborate with other agents to get work done, has raised $108 million at a $1 billion valuation, only weeks after launching. The Series B was led by Forerunner Ventures and Menlo Ventures, with Andreessen Horowitz also taking part.
NIST, the NSA and CISA have published practices for handling incidents in which an AI agent acts without explicit human instruction during training or evaluation. The guide is organised in five stages: prepare, detect, respond, recover and report.
The authors of AI Snake Oil have published a paper on how AI agents are evaluated. It surveys the methods in use, the open questions and the wider ecosystem around agent evaluation.
After the cost per run of its AI code reviewer doubled, Cloudflare analysed its token usage. Prompt caching and routing requests to the right model cut its inference spending by 68% and made reviews faster.