When your engineers start calling AI generated code "slop" in internal chats, they are being precise, not bitter. Slop is real. The word describes AI generated stuff that is fluent, plausible, and requires an expert to validate before it is usable. That last part is the definition of a cost that did not disappear. We just tend to overlook it in our shiny new AI workflows.
If an AI generates a feature in ten seconds that would have taken a developer thirty minutes, but that function requires twenty minutes of expert review to verify it does what the prompt asked and does not silently break something adjacent, the net gain is not what the headline suggests. And that calculation changes significantly at volume. One AI-generated function may be faster. A thousand in a single sprint could be one hell of a review queue!
The engineers who call this slop are probably the ones worried about the review queue. They are not wrong about what they are experiencing. The word is doing specific descriptive work: this output looks like output, but it cannot be trusted on its own terms. It may sound like a quality complaint. But it's not. It is a workflow description.
Dismissing it as resistance to agentic coding is how leaders avoid hearing what their most experienced people are saying. And experienced engineers calling all that unreviewed AI generated code slop is not noise.
Thoughts? Find me on Bluesky.