Environment-Probing Curation Improves Memory for Enterprise Agents

Summary

Persistent memory can help long-horizon agents accumulate experience, but a curator that sees only completed trajectories may preserve errors, overgeneralize from partial evidence, or retain stale information. The paper introduces environment-probing curation, which gives an existing asynchronous curator least-privilege, read-only tools for checking, scoping, and refreshing candidate memories. The deployment-compatible extension requires no model retraining and leaves the task agent, retriever, memory representation, and production write authority unchanged. In a production-like GitHub Copilot harness built with its SDK, the authors compare stateless execution, full in-context learning, GHCP with memory, and GHCP with memory plus environment probing on CLBench database exploration and 90 adapted APEX management-consulting tasks. On CLBench, probing increases the pass rate from 39% to 73% and pass-discounted reward from 8.60 to 22.60, while reducing queries per question from 8.8 to 4.7 and task-agent cost from $3.38 to $1.68. Across six APEX worlds, all 18 memory-versus-baseline mean-reward comparisons are positive, and task-agent tool calls decrease by 16% to 75%. Probing delivers the best task-agent reward gain per dollar in five worlds and achieves higher mean reward than GHCP with memory on both Sonnet 4.6 and Opus 4.7, without schema drift. The authors conclude that environment probing makes memory curation more environment-informed and auditable while preserving a compact task-time interface.