# Okay . AI agents — retained record

> Retained reddit record from reddit:sysadmin.

Canonical: https://highsignal.app/data/records/46efe1b4fddb26ca

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:sysadmin)  published 2026-11-09 00:59 ingested 2026-11-10 02:41

[original source ↗](https://www.reddit.com/r/sysadmin/comments/1xre99t/okay_ai_agents/)

## retained text



```
**Are any sysadmins actually using AI agents for day-to-day administration yet?**  
I know, I know. Security, permissions, hallucinations, and the inevitable story about someone letting an AI agent perform an in-place upgrade on a Windows Server 2008 domain controller.  
I’m not suggesting we hand AI domain admin credentials and tell it to have fun.  
We have Copilot Studio at work, so that’s largely what I’m limited to. But another thread got me thinking about how much of our day is spent doing tedious investigative work that an agent could potentially handle.  
Imagine telling an agent:  
**“Troubleshoot this Wi-Fi failure.”** Collect NPS, certificate authority, Intune, access point, and iPad logs for a specific time window. Correlate timestamps and identify where authentication failed.  
**“Why was this user locked out yesterday?”** Search domain controllers, correlate relevant events, and identify the likely originating device.  
**“Is this AD group actually being used?”** Examine available authentication, access, and audit data and flag groups that may be candidates for review.  
**“What changed before this application broke?”** Compare recent deployments, Windows events, firewall changes, certificate renewals, and configuration history.  
**“Why can’t this server communicate with that server?”** Gather DNS, routing, firewall, and connectivity diagnostics from both ends.  
**“Investigate this ticket.”** Read the ticket, identify relevant systems, collect evidence, summarize probable causes, and draft a response with supporting logs.  
And here’s the part I find fascinating: some agents can interact with a desktop like a person. Move the mouse, open RDP sessions, navigate consoles, run approved scripts, collect evidence, populate spreadsheets, and draft emails.  
I’m all for **human-in-the-loop approval for anything that changes production**. But for read-only diagnostics? Gathering logs and correlating events across five different systems is arguably one of the most tedious parts of being a sysadmin.  
Obviously, access controls, auditability, and data handling still matter, even for read-only operations.  
**My question: Is anyone actually doing this in production today?**  
Not another chatbot that tells me which PowerShell command to run. I’m talking about an agent that does the investigative legwork across multiple systems and comes back with evidence.  
Are we genuinely close to making the most tedious 50% of troubleshooting disappear, or am I dramatically overestimating what these agents can reliably do?  
Especially interested in hearing from anyone using Copilot Studio, Claude, or other agentic tools in enterprise environments.

Notice all these are read only functions…. 
```



document fetched 2026-11-10 00:36 · canonical https://reddit.com/r/sysadmin/comments/1xre99t/okay_ai_agents
