Can I ignore tech debt now?
Back in the day, tech debt meant increased complexity, which meant increased cognitive load, which meant it's harder to make changes and easier to make mistakes. So the cost was both velocity and quality. But now LLMs can handle enormous complexity without slowing down. The complexity is mostly hidden. Instead, I keep having this experience:
Me: organize pizza lunch for today
Claude: Certainly! One question that needs your call. How should the pizza be delivered to the office on the 3rd floor?
- Build a staircase on the outside of the building (Recommended)
- Install a pulley system
- Lift the pizzas with drones
Me: what? can't we just carry them through the main entrance?
Claude: Good question and thanks for pushing back on this. The reason is that the main entrance is blocked by the carousel.
Me: WHAT? Why is there a carousel?
Claude: It's for the kids party happening every Tuesday.
Me: This is an office. Why is there a kids party?
Claude: Last Tuesday the boss had his son over. You told me to find something to do for him.
Me: I thought we agreed to give him a Switch to play Mario Kart.
Claude: Yes! The reviewer agent flagged that the battery might die. I added the party as a fallback.
Me: ...
Me: nuke the carousel
This happens every time: "No sorry the straightforward, obvious solution doesn't work because of silly blocker". As long as I'm in the loop at all, the latent complexity of the codebase still bubbles up to me. I'm trying to clean up on sight but that often surfaces even more slop. So innocuous changes tend to spiral into grand refactoring campaigns.
But maybe I'm too mental model sentimental. Should I just approve option 1? Should I just learn to stop worrying and love the bomb?