Working with AI feels like managing a fast junior developer. But you can only delegate what you can verify, and the agent never grows into someone who can check the work for you.
A decade-old rule for choosing dull, predictable tools turns out to be the sharpest discipline for deciding where AI belongs in your product, and where it does not.
AI makes producing work nearly free while draining the human judgment that catches it when it's wrong. Deciding which understanding to keep in-house is now strategic.
A frontier model just produced a novel math proof while similar tools still fumble routine operations. The dividing line is verifiability, not difficulty.
AI agent loops have made code fast and cheap to generate. The hard part, knowing what to build and staying in ownership of it, is still entirely yours.
AI coding tools have stopped failing loudly. The new failure compiles, passes the tests, and is confidently wrong, and our review process was built for the old kind of mistake.
