Cut long-session token use ~84% with context editing (automatic tool-result clearing)

Digital Applied·high signal

Context editing automatically strips stale tool results and intermediate cruft from the window during long agent runs. Anthropic's internal evals show it delivering a 29% performance lift on its own, and in a 100-turn web-search eval it reduced token consumption by 84% while enabling workflows that would otherwise fail from context exhaustion. Unlike re-summarizing from scratch, it removes low-signal tokens in place — apply it before reaching for a bigger context window, which only raises the ceiling at which context rot reappears.