I use agents to run my troubleshooting steps while I do other things. The information it collects is valuable. It's just smart enough to understand which flow of the troubleshooting playbook to run, but its explanation of what went wrong and what to fix is generally wrong. Like over 50% of the time.
It's a net time savings for me, but I still have to do all of the thinking work. I 100% agree that it is critical to use deterministic tools whether you are executing them or an AI is orchestrating them.