Reading time:

AI is a dishwasher

AI agents are like dishwashers. Yes, you can probably be more precise and clean the dishes better by hand, some of them might be even not dishwasher-friendly and require that you clean them by hand. But for most cases you just load the dishwasher and turn it on and come back in a few hours.

By "AI agents" I mean particularly LLMs used for code generation like: Cursor, Claude Code, Codex, Windsurf and others.

I've seen many people claim that AI is not good at software development and that you shouldn't use it. I think that's like claiming you should not use a dishwasher, a washing machine, an oven or any other appliances that you have in your house.

"But I can do it faster without AI"

I have said this many times, I can do it faster and I can do it better. Many of the times this is the case in software development. Because we're all highly skilled individuals, we know what we're doing and we can do it well.

I'm not an AI fanatic, but refusing to use a convenient tool just doesn't make sense.

Where objections make sense

I've seen Open Source projects reject AI-generated code: cURL and [Qemu](https://github.com/qemu/qemu/commit/3d40db0efc22520fa6c399cf73960dced423b048) come to mind.

And to be honest, I think they're pretty good reasons.

  • cURL is rejecting it because there's too much poor quality reports.
  • Qemu is rejecting it for licensing reasons.

Learn to use your tools

You can generate a lot of code with AI, but that doesn't mean you should do it. You shouldn't mindlessly give LLMs control of your projects.

When you do laundry you need to correctly use the machine and know how to take care of your clothes based on their fabrics, colors and composition. When you fail to separate your whites from your blacks you screw up your clothes. The same happens to your codebase if you do not keep control of AI Agent outputs.

AI is not really a dishwasher

This post is a simplification, there are many more nuances about the topic. Moral, ethic, legal and political implications about current AI models.

You should use your best judgement to when is appropriate to use it.