ARTFEED — Contemporary Art Intelligence

Statewright: Visual state machines for reliable AI agents

ai-technology · 2026-05-13

Ben Cochran, a Distinguished Engineer with 20+ years in full-stack engineering, DevOps, HPC, and ML at NVIDIA and AMD, has released Statewright, an open-source framework that uses formal state machines to make AI agents more reliable. Instead of relying on massive parameter counts or context windows, Statewright constrains tool and solution spaces with visual state machines. Each state defines which tools the model can access, iteration limits, and valid transitions. The approach uses smaller models (13-20B parameters) to solve real SWE-bench problems. The project is available on GitHub.

Key facts

  • Ben Cochran created Statewright
  • Cochran spent 20+ years in engineering and ML
  • He worked at NVIDIA and AMD
  • Statewright uses formal state machines
  • Each state defines tool access, iterations, and transitions
  • Uses models in 13-20B parameter range
  • Solves real SWE-bench problems
  • Project is open-source on GitHub

Entities

Institutions

  • NVIDIA
  • AMD
  • GitHub

Sources