Linus Torvalds on AI and Programming: The Next (and Final?) Layer of Abstraction
Linus Torvalds, the legendary creator of Linux and Git, has shared his thoughts on the intersection of AI and programming, and it’s fascinating. His perspective, shaped by decades of experience in developing software that powers much of the internet, offers a pragmatic take on how AI will change coding—spoiler: it's already happening. In this post, we'll dive into his insights and what they mean for the future of coding, AI, and the next frontier of automation.
AI as the Next Level of Abstraction in Coding
Torvalds kicked off the conversation by pointing out something crucial: automation has always been a part of coding. From early days of writing machine code to the rise of high-level languages like C, Rust, and Python, every advance in programming has been about building new layers of abstraction. With AI-driven tools, we’re witnessing another shift. Large language models (LLMs) may represent the final abstraction layer, where natural language becomes the new way to write code. Torvalds notes, "We don't write machine code anymore... and this is just the next level of automation."
But here’s the kicker: AI could eventually develop coding languages humans can’t even understand. Machines don’t need high-level programming languages like we do. They could evolve their own syntax—symbols, patterns, or something entirely foreign. This is a future where AI writes for itself, potentially bypassing the human-readable languages we’ve relied on for decades.
We might see a coding language that is completely foreign to us… because the only ones that are really going to have to understand it is the AI.
The Role of AI in Code Reviews: Bugs, Be Gone
While Torvalds agrees that LLMs are already being used to write code, he believes their real power lies in debugging and code reviews. AI is particularly good at catching the stupid, obvious bugs—the kind humans often miss. From off-by-one errors to syntax mismatches, these are areas where LLMs can outperform even seasoned programmers.
However, he remains skeptical about their ability to handle more subtle bugs, especially those tied to business logic or functionality issues. While today's AI tools might flag a syntax error, they might not yet understand why a particular line of code doesn’t align with a project's broader objectives.
LLMs are going to shine at finding obvious bugs, but the more complex ones... well, we'll have to see.
This echoes feedback from developers using AI-powered IDEs like Cursor, which are impressive for quick fixes but still fall short in addressing deeper, more intricate issues.
The Future of Programming: Are We Nearing the End?
Perhaps one of the most provocative ideas Torvalds discussed was the potential end of traditional application development. With AI models growing more capable, why build entire apps when you can simply ask an AI for what you need? He suggests a future where applications themselves become obsolete, and natural language commands replace apps like to-do lists, email, or even complex enterprise systems.
What's the point of building an app when you can just tell the AI what you want, and it gives it to you?
While this sounds futuristic, it’s not as far-fetched as it seems. In the world of AI-powered assistants like ChatGPT, you can already see glimpses of this future—people are using conversational AI to handle tasks that once required dedicated apps.
Is AI the End of Programming Jobs?
Torvalds doesn't buy into the fear that AI will wipe out programming jobs anytime soon. In fact, he sees AI as another tool in the long history of software development tools, akin to compilers or IDEs. While it will reduce the need for some mundane coding tasks, it won’t replace programmers entirely.
Torvalds even welcomes AI-driven tools for things like code refactoring, pattern recognition, and error detection. Much like how compilers revolutionized assembly coding, AI will simply be another step in that progression, automating repetitive tasks and freeing developers to focus on more complex challenges.
We're not writing assembly code anymore. Using smarter tools is just the next inevitable step.
The Hype Cycle: Why He’s Not Worried About AI Hype
One of the funniest moments in the discussion was Torvalds’ take on the hype cycle. He’s seen the hype pendulum swing before—from crypto to cloud-native—and he’s cautiously optimistic about AI. But he’s wary of grand claims that AI will completely replace programmers or that we’re on the verge of a jobless dystopia.
Before AI, it was crypto. Before crypto, it was whatever. Now it's cloud-native. There’s always a grain of reality, but you need to be cynical about the hype.
What Torvalds is advocating for is balance. Yes, AI is powerful and useful, but it’s not a magic bullet that will solve all problems overnight. Instead, it’s another tool—a very powerful one—that can help developers write better code, faster.
The Real Game-Changer: Open Data, Not Just Open Source
In a fascinating twist, Torvalds and his interviewer touched on a topic many overlook—open data. While the open-source movement has democratized access to powerful tools and algorithms, the real challenge now lies in getting diverse and high-quality data to train AI models. Open data may hold the key to the next breakthroughs in AI, enabling more people to build, train, and fine-tune their own models.
Torvalds, however, admitted that open data isn’t his personal focus, but agreed that data access is becoming increasingly important. After all, algorithms are only as good as the data they’re trained on.
AI is Changing Programming—But Not in the Way You Think
Linus Torvalds’ take on AI and coding is refreshingly grounded. He doesn’t believe AI will replace programmers entirely, but he does think it will shift the nature of their work. In the same way that high-level languages abstracted away the complexity of machine code, AI will abstract away many of the repetitive and error-prone tasks that slow down development.
What remains clear is that AI is here to stay. Whether it’s used for code completion, debugging, or potentially even creating entire applications on the fly, it will be a tool that developers rely on more and more. The trick will be to harness its power responsibly and avoid getting swept up in the hype.
