"Traditional technical debt is at least visible to the humans who created it… AI-generated systems create a different kind of debt: debt without authorship."
"[W]ho can [fix a system] when the code was generated at scale [without internal consistency]? In many cases, nobody knows where to start because the system was never designed to be understood by humans. It was designed to be produced quickly."
https://www.infoworld.com/article/4141358/the-ai-coding-hangover.html
The answer is: “don’t do it that way”.
I’m not being glib — there are a lot of ways to use tools wrong, and not so many ways to use them right, and we’re in the “cut twice, measure thrice” stage of things.
It’s awfully easy to let the agents run away, but we’re seeing that there’s a cost to doing so. On the other hand, those costs might be like security — hidden, until disaster strikes.
Consider this article in the context of a URL you posted last week: https://digitaleconomy.stanford.edu/app/uploads/2025/12/CanariesintheCoalMine_Nov25.pdf
One of their observations was that job categories where AI *augmented* tasks did not see the decline in junior employees that job categories where AI *replaced* tasks.
Dan Bornstein
David Mankins