AI Slop Is Changing How Engineers Review Code

AI coding tools can generate thousands of lines of code in minutes, but that output still requires review, and large language models can produce code that appears clean while concealing faulty assumptions, security vulnerabilities, or errors that surface only after deployment, according to IEEE Spectrum.

In a survey of more than 1,100 developers by Sonar, an AI code-verification startup, respondents estimated AI contributed 42 percent of the code they added to shared codebases, while 96 percent did not fully trust its output to work correctly. Thirty-eight percent said reviewing AI-generated code required more effort than reviewing colleagues' code, and 61 percent said AI often produced code that looked correct but was unreliable.

Synthesia, an AI video-generation platform, moved 118 engineers onto AI coding tools such as Claude Code in November 2025. CTO Peter Hill said pull requests rose 120 percent year over year as of August, with 95 percent containing AI-generated code. Duplication is a recurring issue; Hill said Synthesia has found as many as 10 versions of the same function. "I don't know if we ever get to the point where you can truly trust the agentic generation of code," Hill said.

At Bonterra, a nonprofit software provider with about 290 engineers, CTO Tanuja Korlepra said proposed changes tripled within three months of AI adoption, code entering review rose tenfold, and review times tripled. Its agents compare code against approved design, security rules, coding standards, and accessibility requirements; low-confidence or flagged changes go to a person, and payments or personal data always get human review. "Agents do the reading and humans do the judging," Korlepra said.

Temporal CEO Samar Abbas said the open-source developer platform's "Send Back" policy requires engineers to explain their agent's design choices and how code handles unusual conditions, or reviewers reject it. "We refuse to let code review become a dumping ground for unchecked model outputs," Abbas said.

At Amazon Web Services, senior principal engineer David Yanacek said agents test whether code works, check it against the original plan, and look for security flaws before human review. McLaren Stanley, a senior principal engineer at Amazon Stores, described a missing instruction that caused an agent to generate 25,000 lines in the wrong version of Swift, producing 600 errors; after he updated the specification, the agent regenerated the code correctly 15 minutes later.

IBM general manager of automation and AI Neel Sundaresan said recent graduates now work on projects once reserved for senior engineers, with AI assisting implementation and testing. He estimated AI can help junior engineers perform 70 to 80 percent of some tasks that once required a senior engineer.