Rauch says AI slop risks people discounting reading itself
The Vercel CEO says the damage from slop is not just bad code, it is that nobody trusts text anymore.
Guillermo Rauch says the real casualty of AI slop might be reading. In a post titled "Reject non-understanding", he argues that slop grenades are not just code and pull requests, and that there is a real risk reading gets entirely discounted because of the exhaustion of having low-quality, unverified AI prose thrown at you all the time.
I want AI in the service of understanding the universe and enhancing human cognition and creativity.
Slop grenades are not just code and PRs.
His example is specific. A thread was going viral crediting a performance improvement to a compiler change. Rauch says the PR description itself contains the AI telling the author that the win was not from the compiler change, it came from how the algorithms and data structures had changed, with the implication that a similar gain could have been achieved another way. So the claim spreading across timelines was contradicted by the document it was built on, and nobody reading the thread had gone that far down.
Why a founder should care
Our read, and it is a read rather than something Rauch says. If people start treating text as guilty until proven otherwise, that hits the cheapest distribution most small software businesses have. Blog posts, changelogs, launch threads, docs, cold emails, the whole written surface of a one-person company works because a stranger will spend thirty seconds reading it. The thing Rauch is describing is that thirty seconds getting shorter for everyone, including for writing that a human actually thought about.
The flip side is the same fact pointed the other way. In Rauch's example, the correction was sitting in plain sight in the PR description. Checking took reading one more level down, and nobody in the viral chain did it. If that is the norm, then being the person who reads the primary document becomes a small, durable edge, in your own engineering decisions and in what you publish.
What he is not saying
Rauch is not arguing against AI. He ends the post with what he wants instead, which reads less like a complaint about tools and more like a complaint about how people are using them. The distinction he draws is between output you can defend and output you forwarded.
There is one post here, from one person, and no numbers attached to it. Treat it as a well-placed opinion rather than evidence of a trend. But the underlying mechanic is easy to test yourself. The next time a technical claim goes around with a link attached, open the link.