Fetching from the wire…
Agents2026-09-17 · source-backed
arXiv 2609.18128 closes the split between LLM judges that score recorded trajectories after the fact and rule engines that block one action at a time. Required behavior goes in as assume-guarantee contracts in LTL over finite traces, compiled to a DFA used for both jobs. It matches state-of-the-art LLM-judge and rule-based baselines across four benchmarks with deterministic reproducible verdicts and orders-of-magnitude lower per-call latency online. The contract library is maintained separately from the agent's model, so it survives a model swap, which is the operational property I'd care about most.
Each link below shares sources, entities, or timing with this story.
After 20+ years maintaining Paint.NET, Rick Brewster concluded WINE's Direct2D would never be complete enough for what he needed, so the app now carries its own from-scratch reverse-engineered Direct2D implementation. He puts it at 180,000 lines against 700,000 for the rest of...
Allen Bargi's August 15 post hit 302 points arguing that AI collaboration rewards context-sharing, examples, and feedback over precise instruction (Hacker News). The pushback holds that the piece conflates management with leadership. mikeocool calls it "the most low effort ver...
Simon Willison released it August 4, calling it "the most significant new version since the initial launch of the project," which from him is not marketing. The agent-relevant pieces: tools can raise llm.PauseChain to stop for human approval, and chains resume from pending cal...
CCP announced the migration covering code that has run on Stackless 2.7 since 2010. The approach is to run futurize across the codebase and then manually review roughly 20,000 places where Python 2 and 3 behavior diverges, including integer division (Simon Willison). No comple...
Simon Willison highlighted Sophie Alpert's short internal policy on AI-assisted writing, whose load-bearing rule is accountability: "You must stand behind every idea and every sentence in your docs," because you can't excuse a bad passage by blaming the LLM, and doing so waste...
GPT-5.6 Luna went to $0.20 input / $1.20 output per million tokens on July 30. That's an 80% cut. Terra dropped 20%. Luna's input now undercuts Gemini 3.1 Flash-Lite ($0.25/$1.50) and sits at one-fifth of Claude Haiku 4.5's $1 input. Simon Willison covered the announcement and...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.