AI
GRAFT: Execution-Guided Genetic Recombination for Agent Design
Publication: cs.AI updates on arXiv.org
Author: Sevda Ebrahimi, Mohammed Saidul Islam, Xiaodan Zhu, Elham Dolatabadi
Source published: 19/11/2026, 05:00:00 UTC
Briefing created: 19/11/2026, 06:22:19 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 19/11/2026, 05:03:25 UTC.
Briefings
GRAFT is a new evolutionary agent design method that uses execution evidence to guide genetic recombination, improving LLM-based agent performance. Evaluated across six reasoning and programming settings, it achieved a mean test accuracy of 82.0%, outperforming the strongest baseline by 9.8%. The method relies on parent-specific execution evidence rather than aggregate fitness alone, with feedback harnesses reusing bounded samples from fitness evaluation without requiring additional task executions.
Key points
- GRAFT uses execution evidence to guide semantic crossover in evolutionary agent design, improving test accuracy by 9.8% over baselines.
- The method evaluates six reasoning and programming settings using open-weight optimizer and executor models.
Context
This research addresses the limitation of standard evolutionary methods that rely solely on aggregate fitness scores, which obscure instance-level successes and failures. GRAFT introduces execution-guided recombination to improve agent design.
Limitations
Results are from a preprint (arXiv:2611.23509v1) and are subject to future validation. Evaluation was conducted across six settings with open-weight models; specific city, region, or country locations are not identified in the evidence.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 19/11/2026, 06:22:19 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeifau5axhmk4cdr52sxcf2mzlbiwol5ycruw3s3aed5wxcbpfdblom