Agent Plasticity paper: frontier models differ wildly in how efficiently they self-improve from experience
_akhaliq · x · 2026-10-09
A new paper, Agent Plasticity: Measuring Self-Improvement Through Experience, studies how efficiently AI agents learn from experience. It examines self-improving agents that turn past interactions into reusable artifacts (tools, skills, memory) inherited by future instances, and introduces Agent Plasticity — a measure of how efficiently an agent converts learning cost into future held-out performance gains. Across Chess, Go, Hex, and NetHack, the authors find that frontier models show very different improvement trajectories despite comparable learning opportunities; the best-performing agent isn't the most efficient learner; training-distribution gains often transfer only partially out of distribution; and effective self-improvement requires producing useful artifacts and applying them correctly, not just retrieving past experience. Evaluating self-improving agents, they argue, requires measuring both endpoint capability and improvement efficiency/reliability.
More from coding & agent
- Claude Opus 5 nearly triples Qwen's SWE-bench score in open-source RSIGym auto-research env — rohanpaul_ai · 2026-10-09
- Khan Academy launches MCP server letting AI assistants read courses and transcripts key-free — modelcontextprotocol · 2026-10-09
- BAAI's AREX research agent checks answers requirement-by-requirement, hits 82.5% BrowseComp — DeepLearningAI · 2026-10-09
- His personal AI agent now auto-compares construction bids in his inbox — dkundel · 2026-10-09
- His 16-Month-Old Call: Claude Code Shifts Agent Workflows and API Spend to Anthropic — majidmanzarpour · 2026-10-09
- pydantic-ai-go brings Pydantic AI's typed agent loop to Go with ~20 providers — samuelcolvin · 2026-10-09