The best AI developers don't trust AI.

People often ask which AI model I use.

ChatGPT? Claude? Gemini? Codex?

My answer surprises them.

All of them.

Not because I'm indecisive. Because I don't believe any single model should become your only source of truth.

When I'm designing a system, I might brainstorm with ChatGPT. I'll move to Claude to reason through implementation. If I hit a particularly difficult engineering problem, I'll ask Codex for another perspective.

Sometimes I'll even ask one model to critique another.

Not because AI is unreliable. Because engineers shouldn't outsource judgement.

The most valuable thing AI has given me isn't faster code. It's better questions.

"Have you considered this edge case?" "Why this architecture?" "What happens if your assumptions are wrong?"

That's where the real value lives.

AI shouldn't replace thinking. It should provoke better thinking.