Avoiding AI Slop
“AI slop” is the failure mode where an AI cranks out something that runs once but nobody — including you — can maintain, extend, or hand off. It looks like progress and quietly becomes a liability.
What Slop Looks Like
The warning signs are usually visible before anything breaks:
- No structure — everything in one place, nothing separated
- No tests and no docs — no safety net, no explanation
- Copy-pasted duplication — the same logic in five spots
- Code you don’t understand — and files piling up unreviewed
How to Push Back
The habits that keep a build clean are ordinary software discipline, applied consistently:
- Clean structure and readable names
- Small modules that each do one thing
- Tests and a README
- Review every change — so you always understand the codebase
Three Properties
A tool that lasts has three properties — each one is a person being able to do something:
- Maintainable — someone can fix it when it breaks
- Extensible — someone can add to it without rewriting it
- Shareable — someone can pick it up and understand it
Keep a Human Able to Take Over
Never let the codebase outrun your understanding. The agent can write faster than you can read — so the discipline is simple: if you can’t explain it, don’t ship it. A human has to be able to take over at any moment.
Build It
How to implement: after your next agent session, spend ten minutes cleaning — rename unclear things, delete dead code, add one test and a line of docs.
- Weekly AI Tasks tracker — keep the parser, storage, and API in separate, tested modules so you can extend it without fear.
- Personal brand site — clean semantic HTML and a short README mean you (or anyone) can update it in a year; ties back to Module 2’s cost and maintenance.
What you learned
Module 4 taught you to shape and spec a build before writing it — and to keep it clean enough that a human stays in control. From here, Modules 5–10 go deep on building the AI application itself.