State machines for agent logic are extremely slept-on: LEGO bricks agents can rearrange
DavidKPiano · x · 2026-10-19
David Khourshid argues that using state machines for agent logic is severely underrated.
Rather than treating them as static workflows, he sees them as LEGO bricks that agents can autonomously rearrange and improve, in a human-readable way. The alternative — freeform prompt/markdown logic — is like "shaping markdown Play-Doh and hoping it holds." The quoted tweet from Steve Ruiz notes the approach has been running for a month with only two restarts.
More from coding & agent
- Dev burns $100k of Claude Code in 24 hours: cheaper to hire a human? — dscape · 2026-10-19
- Distributed Systems Resilience Explained: Circuit Breakers, Bulkheads and Graceful Degradation — _jaydeepkarale · 2026-10-19
- Verified local LLM engine benchmarks: Qwen3.8-27B up to 540 tok/s on a single RTX 5090 — FeydRowan · 2026-10-19
- Free DIY Continuous Security guide: build SAST, DAST, SCA and secrets scanning with GitHub and free tools — evilsocket · 2026-10-19
- New book frames context engineering as an old discipline applied to partly-prompt AI systems — alex_mikhalev · 2026-10-19
- No paid Notion templates needed: a full second-brain workflow with Claude, Granola and Gmail — rubenhassid · 2026-10-19