Context rot: why Every's head of evals starts a fresh Claude thread for every task
danshipper · x · 2026-10-17
Mike Taylor, head of evals at Every, explains why he keeps memory off in Claude and ChatGPT and starts a fresh thread per task: even within one conversation, stale instructions, earlier mistakes, and buried details cause "context rot." Drawing on his forthcoming book Context Engineering with DSPy and Drew Breunig's "How Long Contexts Fail," he outlines seven ways context goes wrong — starting with distraction, where models trained for strict compliance let old patterns constrain new answers — and argues that context engineering is knowing when to start over without losing useful work.
More from coding & agent
- Backend migration from Python to Go now takes one engineer just one week — rakyll · 2026-10-17
- Audio founder says Claude one-shot decompiled rival algorithms, moat gone — ChrisGPT · 2026-10-17
- Google's playbook for agent evals: four layers of checks beyond the vibe check — rseroter · 2026-10-17
- AgentForesight: a 7B online auditor that catches multi-agent failures mid-run — jiqizhixin · 2026-10-17
- Dev gives his agent an alarm tool to wake him during overnight /goal runs — zack_overflow · 2026-10-17
- Dev shares multi-agent orchestration recipe: 2-3 lead agents with many focused subagents — MoonL88537 · 2026-10-17