Salesforce's DarwinX uses an evolutionary approach to fix self-optimizing agent harnesses
bendee983 · x · 2026-09-18
Salesforce Research released DarwinX, a framework tackling two core failure modes of self-evolving agent harnesses: path dependence (the optimizer gets stuck in suboptimal directions built on prior edits) and cross-task interference (optimizing for one task set degrades accuracy on others).
Instead of one continuously rewritten harness, DarwinX generates variants stored in an archive; promising versions keep evolving while useful discoveries from other branches remain available. As VentureBeat puts it: run rival versions and keep only changes that don't regress already-working tasks.
In experiments, DarwinX improved agent accuracy on industry benchmarks, including a 3.4-point gain on SWE-bench Verified.
Related event: Salesforce Unveils DarwinX: Evolving Agent Harnesses via Natural Selection(2 posts)→
More from coding & agent
- Obsidian Starter Kit v4 ships with 375 AI skills, MCP server to make vaults AI-native — dSebastien · 2026-09-18
- Interactive 404 page built with one prompt and Opus: welcome to design engineering — Tegadesigns · 2026-09-18
- Offline replay of 3,300 production decisions validated Jev routing in an afternoon for 20 cents — mikegiannulis · 2026-09-18
- Same Material, Better Order: Reranking Lifts Correct-Passage Hit Rate to 90% — mikegiannulis · 2026-09-18
- 13 'Please Text Me, No Calls' Messages Misrouted — Small Model Caught All of Them — mikegiannulis · 2026-09-18
- Intent Routing 90.5% vs 83.8%, 'Already Did That' Detection 98.3% vs 96.6% — mikegiannulis · 2026-09-18