Tbh, with the current state of AI you also need to write a million test cases so the agent can measure if it succeeded and didn't cause bad side-effects while at it. Which is kind of tricky when dealing with hardware stuff. They'd probably need to buy thousands of computers with all different hardware to test whether tickets for driver issues/shortcomings have actually been solved.

And with AI agents you oftentimes end up in situations anyway where it "solves" one issue but breaks two other things. So the number of open tickets might go down, or we'd end up causing more issues and tickets by this kind of automation...

I'd say it's tricky to impossible to just make AI close a lot of tickets and actually solve them regarding something like an operating system kernel. There's just too much human labour involved and a lot of edge-cases and hardware stuff to be factored in.