Don't Believe Every Polished AI Explanation
I'm currently reading Kahneman's Thinking, Fast and Slow. I've learned a lot about how our judgments and choices get influenced by small things. Sometimes a little word change leads to a different reading of a whole sentence.
Each chapter makes me notice how those strategies show up around me without me realizing it. One situation that came to mind was working with AI.
The word that kept coming back was manipulator. By that I don't mean the model is bad. AI is an expert at producing text, and convincing text. Even when it makes a mistake, or does something you didn't expect, the output often arrives with a polished explanation of why the change is correct and makes sense. That's the trap.
A beautiful, coherent text doesn't mean the change was the best choice. It leads to the same point people keep repeating now that working with AI feels natural: AI speeds you up, but you still decide.
For toy projects, that rigor is not necessary. When the code can make you, or your clients, lose money, the lens should change. Don't believe every AI output. Test it. Ask for a better, cheaper, or simpler alternative. Compare them. Ask what can go wrong. Train yourself on the trade-offs. Same thread as The Step We Skip When Developing with AI.
I don't know how the future will look. Today, at least for me, the human filter is still an important, and necessary, step.
See you next time!