Field notes on using specialised AI agents to produce code, content and audits: what works, what breaks, and why structure matters more than the model.
