Sysadmins have been automating mundane, routine work for decades. PDQ’s 2026 State of Sysadmin report found that 73% of sysadmins want endpoint management to be mostly or fully automated, while 94% see at least one way AI could improve their work. The same report found that 52% are constantly playing catch-up with technology changes. Sysadmins want to automate, and they see AI’s potential, but they already have a lot to keep up with.
But here’s where it gets nuanced. We’re all using AI. But how much of that daily limit or token load is spent doing ad-hoc requests and research versus real automation? We’re willing to make the bet that most of your automations are still rule-driven…as they should be.
We see the potential. We’re just mindful of the shortcomings.
Let’s go back to what got you into IT in the first place. You were a tinkerer. You loved the satisfaction of figuring something out. The light bulb moments made it worth it. Yet, most of what we spend our time doing today…isn’t that. It’s the repetitive, mundane work that keeps the lights on and our organization out of the news. How do we keep that running and spend more time working on the interesting work?
We’re seeing a few emerging success stories.
Where agentic AI is useful now
Agentic AI takes it a step further, past research and into actually doing the work. Here’s where it’s already earning its keep:
- Troubleshooting and diagnosis: If an application deployed across 100 computers fails on six or seven, agentic AI can review the error logs and tell you why. It picks up the long-tail diagnostic work you might not get to and tells you what to do next. Unless you enjoy reading log outputs with a nice cup of coffee.
- Translating an objective into the steps needed to accomplish it: Describe an objective in plain language, and agentic AI can draw on your existing scripts and documentation to turn it into a PowerShell script or the steps you need to get there. This has proved especially helpful for our admins building custom scanners and extending package deployments.
- Report building: Point an AI tool like Claude at PDQ’s MCP server and ask for a “Patch Tuesday impact” report or a dashboard of every device missing Chrome. It builds the report on the spot, scoped to your own permissions and logged in Connect’s audit trail.
Where agentic AI is headed next
A few more areas are already taking shape:
- Turning outside information into action: When Microsoft announces vulnerabilities on Patch Tuesday, agentic AI can show you which devices are affected and generate a report to review. Hit approve, and it remediates them, collapsing the time between finding a vulnerability and fixing it.
- Integrating endpoint management into your full ops stack: Mapping reports into your BI tool, pushing alerts and proposed actions into Slack, diagnosing from within your ITSM, instead of jumping between five tabs to get the same picture.
The bottleneck has moved from doing to reviewing
As AI absorbs more of the junior-level work, it falls to your most experienced people to catch what it gets wrong: the judgment calls about what will actually work in your environment, including the stuff that’s never been written down because it’s just known.
That’s the catch: you need experienced people to review the work, but AI may be quietly taking away the work that turns junior admins into experienced ones. Add in plain old attrition, and you’re losing the people with the instincts (built over years of firefighting) right when you need them most.
Here’s the tension: The light bulb moments that got you into this are exactly what AI doesn’t have. It doesn’t get curious about why something broke. It just applies a pattern.
So how do you solve for it? Make review the new training ground: Have your junior admins check agentic AI’s proposed fixes before they run, the same way they’d shadow a senior admin today. Keep automating the routine work, but spend the time it buys you on investigating the blips, not just clearing more tickets.
Building IT for what comes next
The same instinct that made you a tinkerer is turning out to be one of the best AI skills there is. You’re used to finding where something breaks and working at it until it does what you need. That instinct works just as well on AI.
AI won’t work alone in IT, and people won’t just work around it either. The systems need to let a person step in at the exact moment that matters. That means every agentic action needs a paper trail: what it changed, why, and what it would do next, so you can check the work after the fact instead of taking it on faith. AI that can’t show its work isn’t ready to do more of it.
