Insights on AI automation, ITSM integration, and the future of IT support.
Can AI IT Support Handle Hardware Tickets? What to Expect
Can AI resolve hardware IT tickets, or is it software-only? Here's an honest breakdown of what AI handles well, what still needs a human, and how screenshot analysis helps either way.
Read article →GPT-4 Vision for IT Support: How AI Reads Error Screenshots
GPT-4 Vision lets AI Tech Pal read error screenshots the same way a human engineer would: identifying error codes, application context, and system state directly from the image. This cuts out the back-and-forth of asking a user to describe what they're seeing, and it works on most common error types, from Windows dialog boxes to browser console errors to failed update screens.
Read article →Every Security Check Passed. OneDrive Still Wouldn't Sync. Here's the Layer Nobody Checks.
Two test laptops. That's all a Windows Autopilot pilot was supposed to be for a mid-sized financial services firm I'll call Meridian Capital (not their real name, details here are anonymized, including domains, GUIDs, and internal names). Provision a couple of Entra-joined devices, confirm the basics work, move on. Instead, it turned into a two-week investigation that ended with an engineer one command away from breaking OneDrive sync for the company's entire existing hybrid fleet. Here's what actually happened, and the one lesson worth taking if you're running a hybrid AD and Entra environment and haven't piloted cloud-native provisioning yet.
Read article →Your AI Agent's Prompt Is Not a Filing Cabinet. Stop Storing Facts In It.
Somewhere in a system prompt right now, there's a line telling an AI agent which repository to use, which API endpoint is current, or which vendor tool is the recommended one. It was true the day someone wrote it. It's probably still true today. The question that matters is whether anyone will remember to change it the day it stops being true, because the agent certainly won't notice on its own. That's the whole problem with putting facts in a prompt. Not that it's wrong on day one. That it's wrong on an unknown future day, silently, with nothing to flag it.
Read article →Should Your AI Agent Be Allowed to Reset a Password? The Permission Model That Actually Matters.
The gut reaction to that question usually goes one of two ways. Either "sure, why not, it's just a password reset," or "absolutely not, that's too risky for AI." Both answers are wrong, because both skip the question that actually decides it: not whether the agent can perform the action, but what happens if the agent gets it wrong. That's a permission-model question, not an approval-workflow question, and most people reach for the wrong one first.
Read article →Two SPF Records Will Quietly Break Your Email. Here Are the Other Four DNS Mistakes That Cause Most Tickets.
Nobody gets a warning when they break DNS. There's no red error message when you enter a record wrong, the change just saves, propagates exactly as entered, and then something else fails somewhere downstream, mail bounces, a site won't verify, a subdomain resolves to nowhere, often hours later and several steps removed from the actual mistake. That gap between "the change saved fine" and "here's what actually broke" is where most DNS issues come from.
Read article →Your Spoke VM Can't Reach the Internet Through the Firewall. Here's the Diagnostic Order That Actually Works.
The ticket looks simple. A VM in a spoke network can't reach the internet, even though it's supposed to route through the hub firewall. Someone checks the NSG. Looks fine. Someone checks the route table. Also looks fine. An hour later, still broken, and now there are three people staring at the same two screens that already looked correct the first time.
Read article →The Azure Landing Zone Prerequisite That Blocks Most First Deployments
You have Global Administrator on the tenant. You have Owner on the subscription. You run the deployment, and it fails. Nothing in the error message tells you why. This is the wall almost every first Azure Landing Zone deployment hits, and it is not in most quick start guides. The cause is a permission model that behaves differently than experienced Azure engineers expect, and the fix takes about five minutes once you know it exists.
Read article →Years in Your IT Engineering Role. Building Agents Wasn't in the JD.
You have been in your IT engineering role for years. Building agents was never in the JD, and now it feels like everyone else got the memo. Here is the practical framework map, the three things that kill agents in production, and why your domain knowledge is the competitive advantage you have not used yet.
Read article →