# Linus Torvalds on AI in Programming: A Tool for Joy and a Challenge for Stability
**The open-source world has never been shy about debating new technologies, and artificial intelligence is no exception. When Linux creator Linus Torvalds shared his thoughts on AI during a recent conversation at Open Source Summit Europe, the response was revealing — and full of nuance. Here’s a deeper look at what he said and why it matters for the future of open-source software.**
—
## A Developer’s Journey: From Simple Machines to Complex Code
Torvalds began his programming journey in the early 1980s, when computers were relatively straightforward and the software landscape was far less crowded. Back then, a developer could reasonably understand what every line of their code was doing, and the programs available for comparison were simple enough to learn from.
Today, that clarity has largely disappeared. Modern software engineering involves layers of abstraction, enormous codebases, and professional-grade tools that can intimidate newcomers. Torvalds acknowledged this shift, noting that it’s become increasingly difficult for small, personal projects to feel meaningful when compared against polished commercial software.
Yet this frustration also opened a door — one that AI walks through with ease.
—
## AI as the “Gateway Drug” to Programming
One of Torvalds’ most striking insights was his description of AI as a “gateway drug” to programming. He shared a personal anecdote about a hobby project: building a guitar pedal with a custom user interface written in C, running on a microcontroller with a small screen. The result, he admitted, looked like something from the 1980s. He wanted it to feel modern, but he didn’t know Java — and his expertise lies in kernels, not in crafting sleek consumer interfaces.
So he turned to AI. He gave it a prompt, received a result, and found it was poorly executed. Then he compared it against his own hand-written C code — correct in logic, but visually unappealing. The exercise reminded him that AI isn’t a replacement for understanding; it’s a companion that can help you reach further than you might alone, even if the first attempts are rough.
For new programmers, this is significant. AI lowers the barrier to entry, allowing beginners to build things, experiment, and — most importantly — feel a sense of accomplishment. That feeling of relevance is what keeps people in programming when the learning curve becomes steep.
—
## The Linux Kernel: Where AI Meets Real Responsibility
Torvalds was careful to draw a line between hobby enjoyment and professional responsibility. AI might help someone build a cool side project, but when it comes to the Linux kernel — a project relied upon by billions of devices worldwide — the stakes are entirely different.
“Think about the fact that you need to be very careful using AI when you’re doing something real and important,” he stated plainly.
The Linux kernel receives contributions from thousands of developers worldwide. Maintainers like Torvalds act as gatekeepers, reviewing patches and deciding what gets merged into the official codebase. When AI-generated patches and bug reports flood in, the human review process can become overwhelmed.
### Sashiko: AI-Powered Code Review in Action
One specific tool discussed during the conversation was Sashiko, an agentic Linux kernel code review system. Its public reviews now appear on the Linux Kernel Mailing List (LKML), and some subsystem maintainers have begun requiring that patches receive an AI review before being accepted.
This doesn’t mean AI replaces human judgment. Torvalds emphasized that not every pull request needs AI review, but the trend suggests that most will eventually require it as a baseline check.
### The Double-Edged Sword
AI code review tools bring genuine benefits to the Linux kernel:
– **Security identification:** AI can spot genuine security vulnerabilities that might have been overlooked for years.
– **Legacy code attention:** It can surface issues in old, no-longer-used drivers that have gone unnoticed in the codebase for extended periods.
– **Consistency checks:** Automated reviews can catch style violations, regression risks, and subtle logic errors faster than a human might.
But the costs are real, too. The sheer volume of AI-generated reports and patches can stress maintainers considerably. Torvalds referenced fabricated bug reports that require significant human effort to disprove, and narrowly targeted patches that fix one symptom without addressing the root cause — what he described as “mindless band-aid kind of patches.”
As he noted, the bottleneck isn’t receiving patches anymore. It’s processing them through useful, meaningful review without burning out the people responsible for maintaining the code.
—
## Linux’s Development Evolution: From One Person to a Global Community
The conversation also touched on how Linux development has transformed since its inception in 1991 — a fitting topic given that 2026 marked the project’s 35th anniversary.
### The Early Days: “Me, Myself, and My Computer”
In the beginning, there was no formal development infrastructure. Torvalds tracked changes through tarball patches, distributing updated source code on a regular — initially weekly — basis. He made patches internally every single day and released them just as frequently. For a small project, this approach worked remarkably well.
What Torvalds valued most was independence. He wanted every developer to have their own local setup and environment, working without dependence on a central authority. This philosophy shaped not only how Linux was built but also influenced the tools that would eventually manage it.
### BitKeeper: Controversy and Progress
By the early 2000s, Linux had grown beyond what manual patch management could sustain. Torvalds adopted BitKeeper, a distributed version control system. Its proprietary licensing caused significant controversy within the community — some developers refused to use it. But Torvalds saw it as a success. It demonstrated the power of distributed source control and made it far easier to merge work across subsystems like networking and ARM architecture.
He was also realistic: BitKeeper wasn’t perfect. It showed him what was possible and clarified what he actually needed. Importantly, it didn’t force every maintainer to adopt the same tool.
### Git: The Unplanned Revolution
When BitKeeper’s free use was revoked in 2005, Torvalds needed a replacement — and fast. In just 11 days, he produced the first working version of Git: a new object model, functional code, and a usable tool. The early versions were rough, and developers accustomed to CVS found the new system unfamiliar and difficult.
Today, Git is the de facto standard for version control, used by nearly 94% of developers according to a 2022 Stack Overflow survey. Nearly 97% of professional developers rely on it. Yet Torvalds humbly noted, “I sometimes get way too much credit for Git.” After roughly six months, he handed over maintenance, trusting that the community would carry it forward — and they did, improving it dramatically over the decades since.
—
## The Release Model: Small Changes, Steady Progress
Linux’s release process has also evolved significantly. The project moved away from long-lived, separate stable and development trees toward a model built on frequent releases, merge windows, and release candidates.
Under the old system, work piled up in the development tree while distributions backported selected changes into stable kernels. Annual releases could stretch into years, making long-term planning nearly impossible. Torvalds proposed a dramatically shorter cycle, and the response was skepticism — he recalled that people “looked at me like I had grown a third head.”
His initial target of five weeks was intentionally aggressive. The process eventually settled into a more sustainable nine-to-ten-week rhythm. The result, Torvalds said, is a development model that has worked remarkably well for 20 years and counting.
He also firmly rejected the idea that Linux should chase dramatic, headline-grabbing feature launches. “I’m a big believer in the whole incremental small changes that, over time, turn into big features,” he explained. The goal has always been and remains steady, incremental improvement of an essential and stable operating system.
—
## AI and Linux: The Road Ahead
The intersection of AI and Linux development is still young and evolving. Torvalds sees AI as useful for finding and fixing bugs, improving the codebase, and helping new developers discover the joy of coding. But he draws a firm boundary around where AI should be trusted blindly and where human judgment must take the lead.
The Linux Kernel Maintainer Summit has already reflected this tension: three-quarters of one recent discussion day was devoted entirely to making AI generation and review less stressful and more useful for maintainers.
The future, Torvalds suggests, is about balance. AI is here to stay in open-source development, but it must be wielded with care, responsibility, and respect for the human beings who keep projects like Linux running.
—
## Frequently Asked Questions (FAQ)
**Q: Does Linus Torvalds support the use of AI in programming?**
A: Yes, Torvalds expressed genuine enthusiasm for AI as a programming tool, particularly for beginners and hobbyists. He finds that AI makes programming more enjoyable and accessible. However, he stresses that AI should be treated as a tool, not a replacement for understanding.
**Q: What is Sashiko?**
A: Sashiko is an agentic Linux kernel code review system that uses AI to analyze patches and code submissions. Its reviews are now publicly visible on the Linux Kernel Mailing List (LKML), and some subsystem maintainers have begun requiring AI reviews before accepting patches.
**Q: Why does Torvalds caution against AI in kernel development?**
A: The Linux kernel is used in billions of devices, making reliability and correctness critical. AI-generated bug reports can be fabricated, and AI-written patches may fix surface-level symptoms without addressing underlying issues. This can overwhelm maintainers and potentially introduce instability.
**Q: How has Git changed software development?**
A: Git, created by Torvalds in 2005, became the dominant version control system worldwide. Today, nearly 94% of developers use Git according to surveys, and it is considered essential infrastructure for modern software development.
**Q: What does Torvalds mean by AI as a “gateway drug” to programming?**
A: He uses the term to describe how AI can spark interest in programming among beginners by lowering the initial barrier to creating something functional. Once people experience the satisfaction of building with AI’s help, they may become motivated to learn programming more deeply on their own.
**Q: How often does the Linux kernel release new versions?**
A: The current Linux release cycle follows a nine-to-ten-week rhythm, including a merge window for new features, a stabilization period, and release candidates before the final release. This model has been in place for approximately 20 years.
**Q: Did Torvalds create Git just to manage Linux?**
A: Yes. Torvalds has stated that his goal was to create a tool suited for Linux development, not to revolutionize software development worldwide. He expressed surprise at Git’s widespread adoption and often downplays his role in its creation.
—
## Conclusion
Linus Torvalds’ perspective on AI reveals a pragmatic optimism tempered by hard-won experience. He understands that new tools can open doors for beginners, streamline maintenance work, and catch bugs that humans might miss. At the same time, he recognizes that the stakes in kernel development leave no room for blind trust in automated systems.
The story of Linux — from a single developer’s hobby project to the backbone of global computing infrastructure — is a story of incremental improvement, community collaboration, and the willingness to adopt better tools when they emerge. AI fits into that narrative as both an opportunity and a responsibility. The challenge ahead is ensuring that AI serves the project’s stability and security rather than overwhelming the humans who steward it.
As Torvalds himself has made clear, the goal has never been about flashy breakthroughs. It’s about steady progress, small improvements, and building software that simply works — and that’s a standard that AI will need to meet if it’s to earn its place in the Linux development workflow.
Thank you for reading
