Microsoft paper: verify agent memories at write time — pass rate 39% to 73%, cost halved
rohanpaul_ai · x · 2026-09-15
A new Microsoft paper recommends that long-running agents verify a lesson is correct and reusable before writing it to persistent memory — verification should happen at memory-write time. A finished agent run isn't ground truth: it may contain wrong assumptions, incomplete procedures, or facts that go stale. Their fix: after each task, a separate memory agent with read-only environment access checks what's worth keeping before writing. On CLBench this raised pass rate from 39% to 73%, cut queries from 8.8 to 4.7 per task, and reduced task-agent cost from $3.38 to $1.68.
More from coding & agent
- Cloudflare ships granular authz for Workers, granting agents access to a single Worker — dinasaur_404 · 2026-09-15
- Cloudflare ships granular authz for Workers, its most requested feature — dinasaur_404 · 2026-09-15
- px0 adds millisecond full-text search, project-wide file open and auto-reload — arpit_bhayani · 2026-09-15
- Cloudflare adds granular authz to Workers with four roles for teammates and agents — dinasaur_404 · 2026-09-15
- Pick one coding agent harness, learn it deeply, and stop switching tools — iannuttall · 2026-09-15
- Hypothesis creator hiring London engineers for Hegel, open-source property-based testing — xuanalogue · 2026-09-15