Most of the developers I have known use AI. Some of them use it to build entire apps or to quickly build quick proof-of-concepts. Others refuse to even use it. I’m in the middle myself. I use AI to help move things forward especially when I’m stuck or just want a shortcut through boilerplate, but around 20 years of experience makes it hard for me to trust it.

I still think of code like art, I have my own style and my own way of doing things and no amount of AI generated code will make me change my mind on how to tackle common tasks I’ve been tackling for years.

I’ve been writing code long enough to have strong opinions about structure, style, and behavior. And AI? It tends to have opinions too. Sometimes they align with mine. Sometimes they don’t.

That said, there are a lot of ways to use AI that don’t involve handing over your entire codebase and hoping for the best. These are practical, low-risk prompts that can genuinely save you time:

• “What’s wrong with this code? I expect it to output <expected output>, but it’s giving me <actual output>.”
  Great for catching small logic bugs or syntax errors that your eyes are tired of seeing.
• “Can you generate me an array in <language> consisting of all <thing>?”
  Think: countries, currency codes, HTTP status codes, CSS color names, common file extensions, language keywords. All the stuff you’d normally Google and copy from StackOverflow.
• “How do I do <thing> in Git?”
  Like: “How do I undo my last commit?” or “How do I roll back to a previous version but keep my changes?” Git can be arcane—AI can shortcut the search.
• “Can you generate a regex that matches <string>?”
  Regex is one of those things even senior devs rarely write from scratch. Let AI give you a starting point (and then tweak it yourself, obviously).
• “What does this error mean?”
  Feed it: "TypeError: cannot read properties of undefined" plus a snippet of code. Often helpful, sometimes hilariously off. But worth a shot.
• “Give me a UUID.”
  Because, yes, sometimes that’s all you need.

The way people use AI runs the gamut—some treat it like a coding buddy, others like a Swiss Army knife. There’s no one right way. But regardless of where you land on that spectrum, the fact is: it’s here. And unlike other fads in tech (looking at you, NFTs-for-everything), this one’s not going anywhere.

What I really want to highlight is this: you still need someone technical to guide AI. It doesn’t matter how advanced the model is—AI still makes questionable decisions, forgets context mid-conversation, or conveniently ignores instructions unless you constantly remind it. I’ve lost count of how many times I’ve been working with an AI assistant and it straight-up removed working sections of code I wrote, or even ones it wrote five minutes earlier.

We’re well into 2025 now, and yes, it’s a fun time to be building things. AI is accelerating a lot of the boring parts. But it’s also a tool that needs oversight, context, and sometimes, a bit of skepticism.

Blind trust in AI is a shortcut to tech debt.