What does adversarial use look like? You treat the AI output as a first draft from a smart-but-overconfident junior engineer. You don’t reject it reflexively and you don’t accept it wholesale. You interrogate it.
Here’s a prompt pattern I’ve baked into my actual workflow:Here's the solution you proposed: [paste output] Now argue against it. What are the edge cases this doesn't handle? What assumptions did you make that might not hold in a production system? What would you change if you knew this code would be read by a senior engineer in a security audit?
Run that after any non-trivial AI-generated solution. What comes back is almost always useful — missed error states, implicit assumptions about input shape, security surface area that got glossed over. And critically: you are now thinking alongside the tool, not just consuming its output.
That loop — generate, interrogate, revise — is where judgment lives. It’s where you stay sharp.