Checksum Studio

Agent workflows: Claude Code, Codex, LangGraph. Python. Gated pipelines: deterministic checks first, cheap verifiers second, humans where judgment matters. Generalist with a verification bias.

How verifiers get gamed by the agents they check — adversarial verification patterns for AI agent fleets. Also: making handwritten-receipt extraction trustworthy enough for a real shop.

agent-verification-rails (MIT) — gates that name what they hunt, verdicts that can't be laundered. Small verified automations for non-technical business owners. Writing at checksumstudio.com.

Talking about AI agents that fail confidently — silent failures, laundered verdicts, and the gates that catch them. Say hey if an agent ever shipped something that looked right and wasn't.