Comment by rezonant
9 hours ago
It seems AI code is producing technical debt at an alarming speed. What many people think of as "AIs don't need code to be pretty" is misunderstanding the purpose of refactoring, code reuse, and architectural patterns that AIs appear to skip or misunderstand with regularity. A reckoning will come when the tech debt needs to be paid and the AIs are going to be unable to pay it, the same way it happens when humans produce technical debt at a high rate and do not address it in a timely manner.
I agree with this take. The AI is producing tons of debt still, we will see if that pattern holds or if people automate that part into the agents as well.
Actually, AI's do need code to be pretty. It's becoming widely accepted that whatever is good for humans -- modular code, tests, docs, linters, fast feedback loops -- is good for agents.
I keep repeating this, but the latest data from large-scale dev reports like DORA 2025 and DX find that AI is simply an amplifier of your engineering culture: Teams with strong software engineering discipline enjoy increased velocity with fewer outages, whereas teams with weak discipline suffer more outages.
About the only thing they need (for now) is architectural guidance and spot-checking of results. But then how many architects does any given company need?