August 3, 2026
AI-Written Code Is Now Mainstream. Is Your IT Ready?
August 3, 2026
Linus Torvalds spent years being openly dismissive of AI hype. In 2024 he called it "90 percent marketing." Now he's defending the use of AI-generated code in the Linux kernel. That's not a small thing. The guy is famously blunt and doesn't soften his opinions for anyone.
Around the same time, Microsoft disclosed that somewhere between 20 and 30 percent of its codebase is now written by AI. That number has almost certainly grown since. Windows 11 is also reportedly shipping fewer broken updates lately, which some people are attributing, at least partly, to AI-assisted testing and review.
So what does any of this mean if you're running a 40-person accounting firm in Hoboken or a distribution company in Edison? More than you might think.
AI-generated code isn't inherently bad. In a lot of cases it's faster to produce and catches certain classes of bugs earlier. But it also introduces new failure modes that are harder to predict. The code looks correct. It passes basic tests. And then it breaks in some specific edge case that nobody thought to check.
We've already seen this pattern with Windows updates over the past few years. A patch ships, it works fine on most machines, and then a subset of users with a particular hardware configuration or third-party app suddenly can't print, or their VPN drops, or their Outlook profile corrupts. Those aren't random flukes. They're the kind of issues that come from code being reviewed at scale without enough hands-on testing in diverse real-world environments.
The good news is Microsoft has been getting better at catching these problems before they hit broad deployment. The bad news is "better" doesn't mean "perfect," and your business is still on the receiving end when something goes wrong.
If you're still letting Windows update itself automatically on every machine the moment patches drop, that's a risk you're taking on without realizing it. A single bad update can knock out a critical application for hours. For a small business, that's not an abstract IT problem. That's lost billable time, frustrated staff, and potentially a missed deadline.
The standard approach for businesses our size is to use Windows Update for Business or Intune to create a ring-based deployment. Basically, you let updates hit a small group of test machines first, wait a few days, and only push to everyone else once you've confirmed nothing's broken. It's not complicated to set up, but it does require someone to actually manage it and monitor the results.
Microsoft 365 Apps can be managed the same way. Monthly Enterprise Channel gives you a predictable update cadence with a built-in delay from the latest release, which adds a useful buffer.
AI is now baked into the software you already use, whether you've opted in or not. Copilot features are showing up in Windows 11 and Microsoft 365. The code running your operating system is increasingly AI-assisted. That's just the reality of where software development is heading.
None of that means you need to panic or make dramatic changes. It does mean that "set it and forget it" IT is a worse idea now than it was five years ago. The pace of change is faster, the failure modes are less predictable, and the stakes for a small business when something breaks are high relative to the IT resources most small businesses have on hand.
Keeping a real human eye on your update environment, your endpoint health, and your backup status isn't optional anymore. It's basic risk management.
If you're not sure whether your current setup gives you that visibility, Exine works with NJ and NYC businesses every day on exactly this kind of thing, and we're happy to take a look.
