Paper · Research automation
FreeEvolve
FreeEvolve: Learning to Evolve Beyond Fixed LoopsA framework that automates the optimization loop for agent workflows and uses meta-evolution to improve its own evolution skill.
The loop
The FREEEVOLVE agent automates the design of workflows and evaluation loops for language model agents. It improves its own evolution skill through meta-evolution by scoring each candidate skill on the fresh target agent it produces.
The loop
FreeEvolve- Agent evolver automates workflow design
- Scores candidate skills on target agents
- Improves its own evolution skill
- Produces better target agents
- Agent evolver automates workflow design
- Scores candidate skills on target agents
- Improves its own evolution skill
- Produces better target agents
↻ The improved system does the next round, and the loop turns again.
Why it is a road to recursion
Automating the optimization loop itself removes the human bottleneck in designing how agents are evaluated and improved.
Evidence
Improves the primary held-out metric by 13.6 points on average across tau3-bench, ARC-AGI-2, ARC-AGI-3 and Terminal-Bench 2.1.
meta-evolutionagent-workflowsoptimization
Related loops
More research automation →
autoresearch
- Coding agent edits LLM training script
- Runs a 5-minute training job
- Keeps change if validation improves
- Changes transferred to larger models
- repeat
Research automationAndrej Karpathy · 2026
autoresearch-distillation
- Qwen3-14B edits GPT training script
- Edit is trained and scored
- Score rewards Qwen3-14B weights update
- Trained checkpoint re-run in autoresearch loop
- repeat
Research automationExperiential Labs (Naihin, Fallah) · 2026
DeepScientist
- LLM agents propose hypotheses on AI tasks
- Agents implement and test them
- Findings Memory steers later proposals
- Validated finding directly improves AI system
- repeat
Research automationWeng et al. (Westlake University) · 2025