In this article, I'm going to help you solve the slop problem by encouraging your AI agent to think in tracer bullets, small pieces of functionality that get built end-to-end.
It's a classic software technique that works incredibly well with AI.
AI has a natural inclination to sycophancy. It aims to please, in all aspects of its behavior. "You're absolutely right!".
In code, this means it wants to produce complete solutions all at once. It has in mind the idea for a finished file, or a finished feature, and it produces all of the code needed in one leap.
It doesn't stop to validate assumptions or get feedback. It just keeps going, building entire layers in isolation, without ever testing whether the critical path actually works.
The result? You end up with enormous chunks of code that need reworking. Slop. And there is a huge review burden on the developer to check all this crap.
For example, you ask the AI to build a database service with an API that connects to it.
The AI builds:
GET, POST, PUT, DELETE)Only after all that does it try to connect to the database. And surprise, the connection string format is wrong. Or it's using an incorrect column type. Or the tests don't actually hit the database.
The pragmatic programmer calls this "outrunning your headlights.". The AI is building too much in the dark, without feedback loops to validate assumptions early. And you, the developer, end up with a huge mess to clean up.