I’m frequently asked whether AI is going to replace IT jobs. I’ll let the academics pontificate on AI’s impact on the broader job market, but I believe IT jobs are largely safe. The story in IT is less about replacement and more about reinvention. IT work is simply different now.

Based on everything that I’ve seen across the industry, AI is not replacing IT. It’s just shifting what IT professionals spend their time doing.

Humans in the Loop

Humans will always want humans to ensure that systems are performing correctly. AI is not going to replace that.

Most companies tier their security departments by experience: Level 1, Level 2, and Level 3 triage. The people at Level 3 get paid more because they’re more experienced. Some companies have talked about using AI to automate Level 1 triage. While I think that will probably happen, the people in Level 1 triage are often just beginning their IT careers. Most of their day is spent staring at dashboards and working through repetitive alerts. That’s hardly an engaging way to start a career in technology, nor is it the best use of people who could be developing deeper technical expertise.

This is boring and repetitive work, and it should go away. Earlier this year, IDC reported that AI tools can save IT workers up to 45% of their typical workday by automating routine tasks.

When I worked at Amazon, I appreciated the company’s philosophy: eliminate dangerous and repetitive work that no human wants to do by applying automation. Stacking shelves, picking stuff off shelves, and other manual labor in fulfillment centers is dangerous, repetitive, and boring. Humans don’t really engage in that kind of work.

I think the same is true for jobs in IT. Troubleshooting a high-severity alert is not dangerous in that you’re not going to lose a limb if you misdiagnose something, but you can certainly cause pain to the organization. Take the boring, dangerous, and risk-filled tasks, or the work that computers are good at, and apply massive computational horsepower to fixing the repetitive and error-prone work. We can do all that without replacing jobs.

For example, AI is great at alert correlation (I’m not surprised that Deloitte’s 2026 AI report discovered new roles like “AI operations manager” and “human-AI interaction specialist”). If you’re in IT, you should not be manually trying to figure out whether something happened at this UTC timestamp on this device in London to cause these downstream effects on people’s mobile phone experience in Beijing. If you can figure that out quicker using alert correlation and diagnostics tools, do that.

Tools Will Help Humans, Not Replace Them

There is, however, a trade-off. Across security, IT, and site reliability, companies rely on dozens of tools (Prophet AI found that large enterprises with at least 20,000 employees deploy an average of 28 alert-generating security tools). Companies throw all this budget at IT tools, but none do a great job of solving the fundamental problem.

I’m reminded of conversations I’ve had with customers who were running infrastructure for companies born in the cloud. One SVP of Infrastructure didn’t have any of the old kinds of enterprise organizational constructs. He would tell me that his job is to keep $1 trillion in currency moving around the world, and that he wanted a common toolset that let security, IT, and engineering teams diagnose issues in real time and fix them as fast as possible.

To do that, you need to track the performance of every packet going in and out of the NIC on the individual machine so that, for example, you can check how an application functions on a mobile device for an end user in Timbuktu. That also means being able to track all the transient internet weather patterns in between. The only way to manage such a level of scale and complexity is to use AI to automate the analysis, diagnostics, and root-cause work.

To be clear, I don’t think you can fully remove the grunt work from IT. We will always need humans to parse these AI-generated diagnostics and determine whether the conclusion is accurate. Still, we can remove a lot of frontline support employees’ boring, repetitive work. We can make their jobs a hell of a lot easier on a day-to-day basis.

No Amount of AI Can Stop Outages

Humans make mistakes, and so do AI agents. When you’re running IT services at a global scale, whether for a cloud provider, a bank, or a healthcare organization, the systems are complex and sometimes fragile, with dependencies deep in the stack that are difficult to fully model or understand.

When I was at AWS, I sat in operational meetings where we traced the causes of outages, like someone adding an exclamation point instead of a semicolon in a config file of a network switch. One of my favorites was an unshielded power cable in a data center in Africa. A rat had penetrated a fence, chewed through the shielding on the cabling, and caused a data center outage because somebody forgot to put a 25-cent piece of metal shielding around a conduit.

That’s what I mean when I refer to dealing with weather patterns on the internet. There are high-pressure systems, low-pressure systems, and all the consequences of the fact that we live on a planet that’s circling the sun at over 100,000 km/h. We live in a physical world where stuff happens and breaks. There will always be rain, fire, tornadoes, hurricanes, and so on. Earlier this year, a raccoon caused a power outage in Canada. Fishing ships frequently trawl underwater fiber optic cables. Whether these are accidental or sabotage doesn’t matter to your IT team tasked with getting your business back on track.

IT systems are not immune to any of this reality, and AI can’t prevent any of it. You’re never going to avoid outages, and you need humans to react in real time. When Cockroach Labs surveyed 1,000 senior technology executives worldwide for its State of Resilience 2025 report, 100% said they experienced outage-related revenue loss.

If you go back through the major internet outages of the past year, the next year will likely look similar. There will be outages caused by authoritarian regimes, mobile carriers, and major companies like Cloudflare, Microsoft Azure, AWS, Google Cloud, and even SpaceX. You can’t avoid them, but you can let your AI tools warn you early, so your IT team can reroute traffic accordingly. That way, transient performance issues won’t become a five-alarm fire.