This is my learning blog — a public notebook of things I’m studying and building. Lately that means AI coding agents: what to trust them with, what not to, and the guardrails worth putting underneath them.

Everything I post here is something I actually ran or built; the posts include the commands and their real output.

Things I’ve built

perch
A read-only, local-only, open-source watcher for coding agents. It sits in the macOS menu bar and shows you what your agents are doing — because an agent can’t be the one telling you it’s safe. The thinking behind it is in the essay.
agentpay-proxy
A deployment wrapper for the main agentpay-guard artifact. It pays x402 paywalls while enforcing per-payment ceilings, restart-safe budgets, and duplicate protection below the agent. Guard internals live in agentpay-guard; the walkthrough is the tutorial.

Elsewhere