> When an agent generates a system you didn't write, you lose what cognitive scientists call procedural encoding — the understanding that comes from having debugged line 47 yourself.

How is this different to the person or team who wrote the code leaving the organisation. It's the same thing, its not having the institutional knowledge of the codebase because you can't get it out of the model that made it. So you need to create ways of maintaining that knowledge as its being developed.

Then when you say building the constraint before work begins, don't you mean just understanding the end of result before you build it. Actually identifying functional and business requirements? These aren't anything new in the AI world, they're engineering concepts that stood true before but were often ignored even in an non-AI world.

This article might be helpful for the vibe coder who doesn't have a career in software engineering, but its not news to the field itself.

Coincidentally, tokens are also the currency at arcades. So what's old is new again.

Ai slop eating its own tail.