A developer describes an environment for safely using AI to write Solana programs, where "the cost of a bug isn’t a 500 error" because programs hold user funds and attackers read bytecode immediately. The solution is to make safety "a property of the system" rather than the author. The first guardrail is built on Pinocchio, Anza's zero-dependency framework for SVM programs, which provides an entrypoint and `AccountView` without a macro layer where "subtle behavior can hide." This yields tiny binaries and low compute-unit costs, but requires the developer to write every safety check explicitly. The entire repository is operated through `just`, providing "one discoverable command map" for tasks like `rust fmt-check`, `solana build`, and `creusot verify`. This matters for an AI agent because "an agent's effectiveness is bounded by the quality of its feedback loop"—if "run the tests" means different commands in different contexts, the agent will eventually run the wrong one. In this repository, there is "exactly one way to do anything," listed by `just --list`, and recipes are deliberately thin, declaring a tooling image and shelling into a wrapper.