The Ironies of Using AI Code Generation
Lessons learned from using AI assistants for code generation—why you still need to learn the domain and maintain agency over decisions.
AI code generation is powerful—but only when you maintain agency. After extensive use of Claude, Gemini, and Copilot, I’ve learned that AI is an assistant requiring constant validation, not a replacement for understanding.
Here are the ironies I’ve encountered:
1) I Need to Learn How to Make AI Learn How to Help Me
This is obvious, since we all know prompts are not magic. But I was initially overly optimistic about how much AI already knows or could deduce. It has zero understanding of context, principles, or change management. AI does seem to do much better when working with languages which are already highly formed and disciplined (like backend languages) but when working for front-end applications it is a complete toss up as to what mind bending solution it is going to invent. But who can blame it?!
2) I Am Paying to Contribute to the Service I Am Paying For
I am not certain as to what exact degree our conversations and feedback have on training, but I feel like I have to explain A LOT. This is especially true when trying to leverage a package or API released within the last year or so. If I have to tell the AI that there are 5 bugs in the 100 lines of code that it swears is the best answer then I feel like I am an employee, not a customer.
3) I Am Using AI to Discover What Not to Do
In some cases, not all, AI simply picks a direction and runs with it. It might try to explain why, but given even a slight nudge for it to reconsider it will change its mind even on some of the basic premises of its original. Probabilities are one thing, but randomness is another. Eventually I go back to the documentation for an API, package, or language and then tell AI what I learned.
4) I Spent More Time in Order to Save Time
When I first started using AI code generation I lost so much time. I assumed too much of AI, and I thought that actually learning the “old way” would be slower and less effective than asking my assistant. We learn and AI learns directly from us. So… it does not take a genius to see where this is going.
The Bottom Line
Use AI to assist you. You really should. But you still need to learn the domain and the tech and to have the agency to make final decisions.
Found this helpful? Share it with your network:
Share on LinkedIn