AI helps Samsung design processors in days — but with errors that humans catch

A Kazakhstani outlet retold a telling story: Samsung uses AI tools of the Claude Code class in processor design, and stages that took engineering teams months now pass in mere days. With an important caveat: the generated designs contain errors, and final verification is still done by people — otherwise a defect would reach the silicon, where the cost of a fix is measured in millions.

The story is valuable as an honest measurement of the technology's ceiling. Even in a company with the world's best engineers and a boundless budget, AI does not run unsupervised — it replaces the first eighty percent of the road, not quality control. Speed grew by an order of magnitude; responsibility stayed with humans.

For business in Kazakhstan this is a ready template for treating AI in any task — from code to contracts and calculations. The wrong poles — "AI will do everything itself" and "AI makes mistakes, so we don't need it" — are equally loss-making: the first leads to expensive errors, the second to falling behind those who sped up. The working scheme is a conveyor with control: AI produces the rough mass quickly, humans verify the critical parts. And the higher the cost of error on a given stretch, the stricter the control — as with Samsung and silicon.