Why Some Companies Are Pulling Back on AI Coding

A line of auto-generated code that deletes a customer from a database sounds efficient, even elegant. But what it won't do is archive past invoices for legal compliance, or notify the eight downstream services that need to know the customer is gone.

That gap, between generating code and actually engineering software, is at the heart of a quiet revolt happening inside some of the world's largest tech companies, where AI coding tools are being restricted or outright banned.

These aren't Luddites. They're reacting to real risks: subtle bugs that slip past code review, security vulnerabilities, fragmented codebases, and the uncomfortable question of who's accountable when AI-written code crashes a production system.

Experts estimate we're only 20 to 40 percent of the way toward fully trustworthy autonomous software engineering.

Still, the economics of AI-assisted coding are too powerful to ignore. The likely future isn't one super-model replacing engineers, but AI operating inside tightly validated environments, with humans still holding the wheel.