Exo: an agent harness that improves itself at runtime, slashing cost and time to SOTA solutions
berkeley_ai · x · 2026-11-06
Alex Krentsel (with Berkeley's Ion Stoica amplifying) introduces Exo, a harness architecture for runtime self-improvement. Instead of hand-tuned or offline-optimized harnesses, Exo evolves the harness alongside the solution inside an agentic loop, reportedly bringing big reductions in the cost and time needed to generate state-of-the-art solutions. Paper linked in thread.
Related event: Berkeley's Exo Lets Agents Self-Improve Their Harness at Runtime(3 posts)→
More from coding & agent
- Stable AI proposes DPDM: frozen LLM representations plus shared predictor boost agent decisions — stable-ai · 2026-11-06
- One prompt ran Codex for 17+ hours with 109 sub-agents and 2B tokens — andimarafioti · 2026-11-06
- Reddit devs compare agent memory approaches: summaries, MEMORY.md, wikis and vector DBs all fail differently — ombre-ai · 2026-11-06
- MCP vs API vs CLI: build both on one operations layer, prioritize MCP — hottown · 2026-11-06
- Web Search MCP Server adds Google-powered search and scraping to agents — modelcontextprotocol · 2026-11-06
- DeepMind's Schmid: Build an AI Coworker With 7 Needs Using Gemini Managed Agents — AI Engineer · 2026-11-06