AI That Is Both Clever AND Controllable.

Learn to build AI agents that are briefed, scoped, checked, bounded, and improving with every run. The agent itself stays probabilistic; what you build around it is a harness of deterministic parts, checks that pass or fail and limits that fire or do not. The parts are dependable in a way the agent is not; the whole is safer, never infallible, which is why the harness also watches its own controls. That is what controllable means here. You step back from routine inspection only at the pace your evidence earns it, a person stays on everything consequential, and inspection returns whenever the job, the model or the data changes. The course teaches in plain English and demonstrates everything in Claude Code, an agent tool with an enforcement layer; you direct rather than program, and where a control needs one-time technical setup the course says so. It follows one recurring job, a monthly run of financial documents, simplified for teaching and not a template for live accounts processing; the course names what a real process adds, from duplicate checks to separated approval. The documents are only the example: the method is for anyone with a job that repeats and has to come out right.