We all agree that blindly using AI will deteriorate code in the long run at least at this point. It's awesome when you can deliver your craft at the highest level possible. But if you are going to build a treehouse, you don't need the same details and craftsmanship as if you are building a skyscraper.

But at some point, you have to take economics into account. It depends on what you are building. Some code will guide a rocket, some will be a banking system, some an ad builder. Where is the threshold on what is economically the best decision and how much correctness you need, we all land differently on this scale. Are we in a position where we can be slow and correct or do we need to be fast and make mistakes?

At the end it’s not outputs that matter but outcomes.

It’s worth being in the loop and testing the agent with internal tooling helper scripts, etc.