Story
arxiv_cs_cl · Nov 8, 2026 · paper
Source brief
Agent Self-Improvement with Frozen Weights: The Added Value and Limits of Failure Traces
arxiv.orgNov 8, 2026
original source linked
In brief
We evaluate whether an AI agent's harness, the notes around its model, can learn from the agent's failure traces beyond a matched search that never sees them. It can, but less than it looks. We tested an open 9-billio...
Feed lens
agentharnesseval