From the Workshop
On governance, local-first execution, and the architecture behind a runtime for AI agents and MCP tools.
September 17, 2026 · 8 min read
Risk isn't a property of the verb
I pointed an adversarial review at my own agent runtime. It found argument-blind risk evaluation and approvals that allowed self-approval — plus what MachinaOS 1.1 does about both.
July 1, 2026 · 7 min readMost AI agent frameworks have the safety model backwards
Agent frameworks optimize for autonomy and bolt on safety later. That's backwards. Here's the governance model behind MachinaOS — risk-gated execution and a hash-chained audit log.