I recently read “Slop Paralysis” by Elijah Potter (a colleague at Automattic). It’s a useful name for a familiar feeling: moving “fast,” then not wanting to look too closely at the result. I really liked the framing so I took these notes and made a NotebookLM 1pic-recap (the featured image).
Has it happened to you that you let an agent work, it generated A LOT of code, and then you get stuck when it’s time to review it? I mean, not out of laziness, but because of the volume, the missing context (An agent tends to use way more context than we are aware of), and the fear of breaking something. And I’d add, the intricacy of the code that follows conventions you didn’t mean to, is overly verbose sometimes, overcomplicated others… You know, the overcomplicated kind when you think it works, but it takes 20 minutes to be sure how.
The short takeaway from the post:
- Don’t use an agent by default.
- Ask the agent for a plan before asking for code.
- If the solution is bloated, trim it hard.
- Refactoring by hand helps recover real understanding.
Leave a Reply