Fixing Coding Agent Amnesia: Graph-Managed State Substrate Benchmarks
ShanRizvi · x · 2026-07-29
To solve the "amnesia" problem where AI coding agents lose objectives and lessons upon restart, a team launched a graph-managed work state substrate for Claude Code and Codex. This mechanism turns tasks, iterations, and lessons into durable graph state, allowing fresh sessions to resume seamlessly with a single bounded brief.
The team provided core performance benchmarks:
- Accuracy: 20/20 correct work selection across a scripted resume matrix (structural selection, not embedding-ranked)
- Speed: p95 brief injection at 133ms on a local server
- Concurrency: Zero duplicate identities under a 20-way concurrent anchor race
Related event: MindGraph: Persistent Graph Memory for Coding Agents(8 posts)→
More from coding & agent
- Harness open-sources desktop and mobile apps for orchestrating agent fleets across frameworks and machines — dee_hw · 2026-10-16
- Open-Source Google Analytics MCP Server Lets LLMs Query GA Data via Natural Language — modelcontextprotocol · 2026-10-16
- LLM ideation experiment: just ask for non-obvious ideas — and Opus beats smaller models by a wide margin — arjunrajlab · 2026-10-16
- a16z podcast: AI cut a 6-month feature to 10 days and widened engineer gaps 10x — a16z · 2026-10-16
- "Hold On, My Agents Are Freezing My Screen": The Daily Grind of the Agent Era — every · 2026-10-16
- Dev builds MCP feeding Claude 3,000 real launch videos to fix samey AI output — Designer_Landscape25 · 2026-10-16