I spent years as a programmer the traditional way: specs, code, debug, repeat. When I started building with AI, the first lesson wasn't about prompts. It was about trust.
Early on, when the AI raised a concern about my code, I read it as a sign that the tool couldn't be trusted. I had it backwards. A concern is not a defect. It's a question, and it's the first change every programmer has to make when working with AI.
What changed my results was discipline, not cleverness:
- Define the job completely, including what the project will not do. Never start work without a complete definition.
- Build trust through proof. Use the AI to define the tests up front, so every result is verified and not just taken on faith.
- Design well-defined parts. Normalized data and clean APIs make systems that AI can build, and that people can extend.
Programs are deterministic, so AI-built software can be proven. That's very different from using AI for research or polling, where you sift for the pearls. Knowing which game you're playing is half the work.
I wrote up the full set of thinking tools in an essay, A concern is not a defect.
Read the essay at thinking.bongo.cc →
Tell me about your challenge
Stuck on something similar, or have a project in mind? Tell me what you're up against and I'll get back to you.