Grounding Agent Memory: Environment-Probing Curation for Enterprise Agents
Susheel Suresh and colleagues at Microsoft give the memory-curator agent in a GitHub Copilot harness read-only tools to check candidate memories against the live environment before they are saved, which roughly doubles pass rate on a database exploration benchmark and halves cost.
Ask this paper
Problem: A post-task curator that only sees completed trajectories can save errors, overgeneralize partial evidence and keep stale knowledge.
Environment-probing curation: The existing asynchronous curator gets least-privilege, read-only world tools to verify, scope and refresh memories. The task agent, retriever, memory format and write authority stay unchanged, and no retraining is needed.
CLBench results: Pass rate rises from 39% to 73%, pass-discounted reward from 8.60 to 22.60, queries fall from 8.8 to 4.7 per question, and task-agent cost drops from $3.38 to $1.68.
APEX consulting tasks: Across six worlds and 90 tasks, all 18 memory-versus-baseline reward comparisons are positive, tool calls fall 16 to 75%, and probing gives the best reward gain per dollar in five worlds.
Model coverage: Probing beats plain memory on both Sonnet 4.6 and Opus 4.7 without schema drift.
Abstract
Persistent memory is entering production-oriented agent platforms to help long-horizon agents accumulate experience across sessions. Yet a post-task curator agent restricted to completed trajectories can preserve errors, overgeneralize partial evidence, or retain stale knowledge. We introduce environment-probing curation, a deployment-compatible extension that gives an existing asynchronous curator agent least-privilege, read-only world tools to check, scope, and refresh candidate memories. It requires no model retraining and leaves the task agent, retriever, memory representation, and production write authority unchanged. In a production-like GitHub Copilot (GHCP) harness built on its SDK, we compare stateless execution, full in-context learning, GHCP + Mem, and GHCP + Mem (w/ Env Probing) on CLBench database exploration and 90 adapted APEX management-consulting tasks. On CLBench, probing raises pass rate from 39% to 73% and pass-discounted reward from 8.60 to 22.60 while reducing queries from 8.8 to 4.7 per question 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 fall by 16--75%; probing gives the best task-agent reward gain per dollar in five worlds. Probing also attains higher mean reward than GHCP + Mem on both Sonnet 4.6 and Opus 4.7 without schema drift. Environment probing therefore turns existing agent-memory curation into an environment-informed, auditable process while preserving a compact task-time interface.