Research: Agent Behavior Shaped More by Harness Than Underlying LLM
dair_ai · x · 2026-08-28
New work collapses agent traces into finite-state machines (FSM), showing complex agent behaviors can be represented with just 7-43 states. FSM context outperforms agent workflow memory in next-step prediction and failure prediction (AUROC 0.94). The authors suggest behavioral topology is shaped more by the deployment harness than the LLM itself.
More from coding & agent
- WebMCP Gains Traction: Netlify to Demo Conversational Checkout — thisiskp_ · 2026-08-28
- Agents made 800k+ production writes with zero human approvals, failure rate 28%→6.5% — TimothyDuignan · 2026-08-28
- Rippling launches MCP server: Letting agents execute code instead of calling APIs — andreisavu · 2026-08-28
- Paseo: Open-source agent orchestrator with full mobile support — jasonkneen · 2026-08-28
- LangChain Tutorial: Building Self-Correcting Memory in OpenWiki — adnan_hashmi · 2026-08-28
- Uber Cuts AI Costs by 34% While Scaling Agent Usage 9.4x — dosco · 2026-08-28