ReGE Framework Improves Agent Accuracy via Experience Synthesis
October 23, 2026
The ReGE framework synthesizes recorded execution traces and artifacts into self-contained guidance for new task attempts. Using a single coach call without parameter updates, ReGE-Syn improves agent accuracy by an average of 9.9 percentage points across four benchmarks.
HOW THIS AFFECTS YOU
●
builderYou can improve agent reliability by synthesizing past successes and failures into actionable guidance for new runs.
●
researcherThis demonstrates a method for test-time improvement without requiring teacher model parameter updates.