Is AI slop taking over? Claude Code developer has an answer

A developer wrote to Boris Cherny with a one-line subject: "What to do about slop?" Cherny says he gets versions of that email every single day.

He posted this one on X anyway, along with his full reply. The developer had laid out a divide splitting engineering teams right now and asked Cherny to settle it.

On one side, the code created using AI technology should be reviewed, explainable by anyone who submits it, and as maintainable as before. On the other hand, this is an excessive effort. You should just validate the result and proceed, as the code remains a black box which doesn’t need to be opened.

According to the developer, most of the users cannot implement the first method properly because of the fear of being lagging behind. "A lot of people are just forcing it," he wrote.

Cherny didn't pick a side. Throwaway prototypes, he said, can stay black boxes if the code gets discarded soon and nothing breaks badly when it fails; polish doesn't matter.

In the case of production code, however, things are a bit different. According to Cherny, the code written by Claude requires a stricter set of standards than a human-written code, not less strict standards. This opinion is supported by Anthropic in-house through extensive linting, automated end-to-end testing, daily fuzzers, and continuous security review.

Cherny's framing puts the responsibility for code quality squarely on the person directing the AI engineers, increasingly acting as something closer to a manager than a typist, catching bad output before it ships rather than writing every line themselves.