Show HN: Statewright – Visual state machines that make AI agents reliable
Smaller models. Fewer tokens. More reliability. Ben Cochran just shipped Statewright—a visual state machine engine that makes AI agents actually work without scaling to frontier models.
Why it matters
A working engineer from NVIDIA/AMD just dropped a practical fix to agentic brittleness: constrain the solution space via formal state machines instead of scaling model size. Early results show 13-20B models outperforming larger ones on SWE-bench when paired with Statewright's deterministic orchestration. This is a product play that challenges the 'bigger models = better agents' orthodoxy.
The key facts
8 to knowStatewright uses Rust state machine engine to enforce tool/step constraints per agent state
Works with models in 13-20B parameter range; improvements consistent across Qwen, GPT-OSS, Gemma4 families
Smaller models (Haiku, Sonnet) punch above weight; Opus uses fewer tokens with fewer 'death spirals'
Live product with free tier; integrates via MCP plugin with Claude Code, roadmap includes Codex and Cursor
Visual editor shows failure paths, retry loops, and approval gates; state machines enable loops vs DAGs
Core insight: context window utilization matters more than raw context size; tightly-scoped working context outperforms unrestricted access
Founder: Ben Cochran, 20+ years in full-stack, DevOps, HPC & ML; former Distinguished Engineer at NVIDIA/AMD
HN post (May 12 2026): 4 points, 0 comments at time of publish
Go to the source
Hacker Newsgithub.com
Publisher excerpt: Agentic problem solving in its current state is very brittle. I fell in love with it, but it creates as many problems as it solves. I'm Ben Cochran, I spent 20+ years in the trenches with full-stack Engineering, DevOps, high performance computing & ML with stints at NVIDIA, AMD and various other…