Arguing with your AI

The best developers argue with their AI.

I used to be against AI assistance. Now I use it every day, but with a rule: AI speeds up the typing. I make the decisions. Architecture and algorithm choices stay with me. Boilerplate, test scaffolding and debugging help are fair game.

Two ways I’ve used it

In Launch Control and Axis, I wrote the code myself and brought AI in for specific jobs: writing tests, chasing down bugs, and making UI tweaks.

In Drift, I tried the opposite. It’s my first app built entirely by prompting. I came up with the concept and the feel, wrote a detailed design spec (claude.md) that gives the AI full context at the start of every session, set the build order, and reviewed every change on a real device.

Drift taught me that prompting well is mostly engineering judgment. Two examples where I pushed back:

• I dropped MapKit for a custom street-level renderer drawn in a SwiftUI Canvas.
• I replaced placeholder geometry with real OpenStreetMap roads and buildings once the placeholder stopped telling me anything useful.

Neither change would have happened if I’d accepted the first output.

How I argue with it

• Ask “why” before “write.” I ask the model to explain its approach before it writes code. If the reasoning is weak, the code will be too.
• Ask for alternatives. “Give me two other ways to do this and the trade-offs.” The first answer is rarely the only good one.
• Make it show its work in the code. In Drift, every file has doc comments explaining why it works the way it does. That’s the part I learn from.
• Verify on device. Generated code that compiles isn’t the same as code that works. The simulator and a real phone don’t always agree.
• Use it to understand, not to skip understanding. My best sessions are rabbit holes like “walk me through why Swift 6 flags this as a data race.” I come out knowing something, not just having something.

The risk

If you only accept answers, you stop building judgment, and judgment is exactly what makes the AI useful. Arguing with it is how I keep learning while still moving fast.