Linus Torvalds: AI-Assisted Debugging and Thoughts on Qwen 3.8

In a series of messages in August 2026, Linus Torvalds recounts having conducted a challenging debugging session with the support of an AI, which he describes as both helpful and quick to give up. He even mentions having entrusted the tool with drafting a commit message. The developer also evaluates Qwen 3.8 27B and notes the existence of a timeline regarding an incident involving OpenAI and Hugging Face.

Torvalds Reports a Timeline Related to OpenAI and Hugging Face

On August 7, 2026, Linus Torvalds claims that a timeline of OpenAI's accidental attack on Hugging Face is now available. This quote was collected by Simon Willison, who published it on August 22, 2026.

Torvalds Describes AI-Assisted Debugging, Message Drafted by the Tool

On August 22, 2026, Linus Torvalds describes a "hellish" debugging session, largely facilitated by an AI that did much of the work. He adds that the tool repeatedly concluded that the problem was impossible and unsolvable, even recommending to stick with a report. Torvalds suspects that these systems have been trained by people less stubborn than he is. Despite these signs of giving up, he reports that the AI continued to propose and analyze debugging code when he pushed it to do so. He finally indicates that he let the AI draft the associated commit message, in a context related to "drm/xe: Do not expose flat CCS storage as usable VRAM."

Torvalds Evaluates Qwen 3.8 27B and Mentions "Conceptual Integrity"

On August 16, 2026, he judges that "Qwen 3.8 27B is excellent," while pointing out a tendency of this model to "overthink." Three days later, on August 19, 2026, he writes "Conceptual integrity and line counting."