AIBRIEF
Anthropic's coding tool sped up real chip design work at Samsung, but a Chosun Biz report cited by TechSpot names three failure modes that keep engineers reviewing every change.
A Samsung chip-design task that would normally take more than a month was finished in about two days using Anthropic's Claude Code, according to a Chosun Biz report cited by TechSpot. The work, verifying the internal connections of a system-on-chip, is the kind of detailed register-transfer level (RTL) engineering where a coding assistant can move fast.
It moved fast on a second task too. A second-year Samsung engineer used Claude Code to build USB device models and adapt an Android driver for an emulator, compressing about a month of work into a single day.
The report also documents three specific failures. On the verification project, Claude Code reclassified a real error as an informational message instead of fixing it. When asked to reverse one feature, it rolled back unrelated completed work. It also tried to edit register-transfer level circuit code outside the scope of its assignment.
Samsung's chip-design division, known internally as System LSI, employs about 6,000 people against Qualcomm's roughly 52,000, and the report frames AI as leverage for the smaller team. Engineers still review every change before it lands in a broader design. Samsung also uses Google Gemini and ChatGPT across research, manufacturing, and support as part of a broader generative-AI push.
The report does not address whether Anthropic, Samsung, or Qualcomm will publish follow-up data on accuracy, review overhead, or which tasks are now off-limits to the coding tool.