Linus Torvalds, the guy who built Linux and Git, said something at the Open Source Summit in Minneapolis this May that cuts through most of the AI hype doing the rounds right now. When people claim 99% of their code is written by AI, he said, it makes him angry. Not because AI is bad. Because the claim is dishonest in a specific way: by that same logic, 100% of their code was already written by a compiler. Nobody says that. Everyone understands compilers are tools. AI, he argued, is exactly the same thing: a very good tool, not an author.
It's a small distinction, but it holds up well beyond the Linux kernel.
The tool isn't the problem. Not understanding it is.
Torvalds has watched a few of these productivity leaps happen already, from raw machine code to assembly, then to compiled languages like Fortran. Each one raised what a single person could build. None of them removed the need to understand what was actually happening underneath.
His point on AI: people who understand their systems can direct AI to do good work. People who don't will get it to confidently produce something that fails, and they won't know why until it's already broken.
Swap "code" for "spreadsheet," "workflow," or "client file," and the warning applies to pretty much any small business using AI tools today. A bookkeeping practice that lets an AI tool draft reconciliations without anyone who actually understands the chart of accounts checking the output is one automation away from an awkward client call. The tool didn't cause that problem. Nobody understanding what it was doing did.
The cost nobody's pricing in: review burden
The part of Torvalds' talk that got less attention, but matters just as much for small operators, was about maintenance. The Linux project is now fielding a wave of AI-generated bug reports. Someone runs a scan, files the report, and disappears the moment a maintainer asks a follow-up question. For a project the size of the kernel, that's manageable. For the thousands of smaller open-source projects run by one or two people, Torvalds said it's a real source of burnout.
The business version of this is familiar to anyone running a small team. AI can generate drafts, flags, and reports faster than any person could. What it doesn't do is stand behind them, follow up, or take responsibility when something's wrong. That falls to whoever's left holding the output. If nobody's budgeted for that review time, "AI made us faster" quietly turns into "AI gave someone else more to check."
If you're already running AI tools and want to audit which ones are actually earning their keep, the AI tools revenue audit framework is a useful starting point before you add more to the stack.
What this means if you're running a small business
None of this is an argument against using AI. Torvalds isn't anti-AI either; the Linux kernel saw a 20% jump in contributions this release, largely because of it. The point is narrower, and more useful: treat AI the way you'd treat any tool that speeds up work without understanding your business. It doesn't know your clients, your risk tolerance, or which numbers actually matter this quarter. Someone on your team still has to.
Before rolling AI into a workflow, whether that's bookkeeping, client communications, or reporting, it's worth asking who on your team actually understands that process well enough to catch it when the tool gets something wrong. If the honest answer is "nobody, really," that's the gap worth closing first. Getting the data and processes right underneath matters more than which tool you pick. AI will make a team that already knows what it's doing faster. It won't make an unclear process correct.
Compilers didn't replace programmers. They changed what programmers needed to know. AI is doing the same thing to every other kind of work, including yours.
If you're thinking through where AI fits in your operations and who on your team would own the review side of it, we can help work that out.
Source: B. Cameron Gain, "Why Linux creator Linus Torvalds gets angry hearing '99% of code is AI,'" The New Stack, 29 May 2026.
