Most of my work sits where software meets accountability: who can do what, and how you prove what happened. Right now that means payments by day and AI agent guardrails in the open.
Building Nanny: an open-source authorization and audit layer for AI agents. You declare which tools an agent may call; if it tries anything else, Nanny stops the run and records why. Docs
Latest essay: Put the human in the system, not the prompt
evns.fyi · LinkedIn · X · evans@evns.fyi
This repo is also the source of evns.fyi, built with Astro.




