Slop
Slop is the smell of sameness. It is what people start to notice once a tool can produce the same kind of okay output everywhere. You detect it with taste, not a checklist, and the defence is a human-owned core the AI cannot touch.
When yesterday's competence goes cheap, every domain floods with generic output, ranging from complete slop to a decent start. That glut of sameness is exactly what raises the value of difference. Slop is the visible residue of cheap competence, and taste is the thing it makes scarce.
The trap is treating slop as a list of tells, like too many em dashes. It is not mechanical. It is a judgment, which is why you cannot lint your way out of it.
What it says
- Slop is a smell, not a checklist. Dan Shipper: it is not any one thing like too many em dashes. It is what humans notice when a tool can produce the same kind of thing in many circumstances. Everything starts to look a little similar, and you smell a rat.
- The mechanism is cheap competence to glut to a premium on difference. Adoption explodes, the market fills with generic output, and the scarce thing becomes the person who can make something different. Slop and the taste premium are two sides of the same dynamic.
- Every tool ships its own slop. Claire Vo, reviewing Google I/O 2026: every AI tool ships its own version of slop. Google's slop looks like Google. Claude's slop looked like Claude. A model's default aesthetic is a fingerprint, and the more a surface leans on those defaults, the more it converges on that model's slop.
- Slop compounds through feedback loops. In code the danger is mechanical. The AI reads bad code and writes more of it, a vicious cycle, and the same runs in the positive direction. Left unguarded, one batch of slop becomes the context for the next, so quality decays geometrically.
- The defence is a slop-free zone. Charlie Holtz marks a human-owned core the AI may not touch: architecture, core APIs and contracts, and interface decisions. His codebase has lines that say do not touch if you are an AI. His rule: don't let the AI be your architect.
- The expert response is the second pass. When everyone can produce okay, okay stops being valuable. The work is not to do more. It is to instrument the standards that catch slop, then do the human pass machines cannot.
In their words.
"Slop is not any one particular thing. We start to smell a rat. Everything looks a little bit similar." — Dan Shipper
"Every AI tool ships their own version of slop. Google's slop looks like Google. Claude slop looked like Claude." — Claire Vo
"Don't let the AI be your architect." — Charlie Holtz
Worth knowing
What if the market just accepts slop? The whole defence assumes buyers can smell sameness and will pay for difference. In low attention or commodity contexts, good enough may beat actually good. None of the sources quantify how often.
The boundary is murky. Holtz calls his own slop-free line a little murky, and that is exactly where the risk lives. There is no clean rule yet for what stays human-owned versus AI free-rein. It is per-codebase judgment.
Related
- Cheap competence premium — the economic engine. Cheap competence produces the glut whose smell is slop, and raises the premium on difference.
- Taste as bottleneck — taste is both how you detect slop and the scarce input it makes valuable.
- AI evals — evals are the structured way to catch slop at scale instead of relying on a nose for it.
- Brand code — the positive version: give agents a strong substrate so their default output is less generic.
Sources
- AI Was Supposed to Save Time. Why Am I Busier? — Dan Shipper, Every. youtube.com/watch?v=jBZQ5Ay20HU
- Was Anything Good Released at Google I/O 2026? — Claire Vo, How I AI. youtube.com/watch?v=S-T12Z-8Tbo
- Charlie Holtz on His AI Coding Setup — Y Combinator. youtube.com/watch?v=fQmlML9Lay4