Torvalds: AI is fine for competent devs, but spare us the 'drive-by' patches

Linus Torvalds argues that AI is a legitimate programming tool when used by developers who understand the code and can judge its output. Linux kernel maintainers have therefore decided to drop the “Assisted-by” tag, removing the requirement to disclose AI use in submissions. That change does not lower the bar for acceptance: Torvalds wants the community to reject “drive-by” patches from people who generate code without understanding it or intending to learn, while welcoming newcomers who use AI as a route into programming and work to become capable contributors. The key distinction is not whether AI was involved, but whether the person submitting code takes responsibility for its quality.

AI is also creating real work for maintainers by finding genuine bugs across large codebases and generating a surge of reports, though those discoveries may eventually reduce the pool of easily detectable problems. Other practical benefits include tools that automate the investigation of hardware behavior, making it easier to develop Linux drivers from Windows-only drivers and potentially improve support for poorly served devices. The argument acknowledges serious concerns about AI, including weak training data and disregard for copyright, but says these do not negate its usefulness in skilled hands. AI exposes two separate challenges: handling valid findings at scale and filtering uninformed submissions. The broader prediction is that AI will become a standard part of software development, while production-quality work will still depend on human competence.