First off, I’m a very practical kind of person. For the past 20 years, server names have always looked like “Server22,” referring to what it is and what year’s operating system it is running. Easy, descriptive, just a quick and dirty way to know what you’re looking at. I walked into AI Agents the same way. I thought having silly names for them would be distracting and pointless. And maybe I will at some point change these, but for now, I’ve decided to have a bit of fun with the names. And I’ve also decided to build out several of them to help fill in the blanks around here.

First up is Sindri. Sindri is one of the forge masters who created Thor’s hammer, among other things. His role in our organization is to build our internal asset inventory in Microsoft Dynamics. I know, it’s a stretch, but hey, it’s fun. Currently he is hanging out in Microsoft Teams and Copilot, and you can do things like send him a screenshot of a Windows About page. He’ll verify if the asset already exists, and if not, read in the values to create a new asset. It’s rather basic right now, he needs more instructions to get it all in there, but for a version 1 tool, I’m happy with him.

My other agent I’ve been building is Huginn. Huginn is one of Odin’s ravens, and his job is to gather intelligence about sales opportunities from our team and get them documented into Opportunities in Microsoft Dynamics. We can send him information about a potential device replacement, and he’ll make an opportunity for us to start tracking. Soon he’ll be able to also tell us about the progress of an opportunity as our sales team moves it through our system. I just need to build out some more tools for him to use.

I’ve got plans for more agents too! One that I think could be really helpful is Heimdall. Heimdall will be our eyes on all our systems. He’ll read in emails, event logs, alerts, and whatever else we can feed him. Then he’ll decide if these things are important enough for us to generate a support case, raise a big red flag, or just log it for later. We’ll also be able to ask him if there have been issues with a site recently. While we have an actual XDR product called Sonic Sentry that will go deep, Heimdall will be mostly just reading through all the alerts we get to track them.

Finally, Ratatoskr, the squirrel tale bearer (if you ever played Hammer of the Gods, you know). This guy will be our messenger to our clients. I’m planning on him being publicly available on our website, but this is where I need to be very careful. I’m not sure how yet, but I want clients to be able to check on their support cases and update us on the status of things without calling if they want to. So ideally, Ratatoskr would verify your identity and then either put a note on your case for you or give you a quick summary of the last action item on your case. Because of the sensitive nature of our case work, we’ll need to know you are who you say you are before he can talk to you. I’m not sure how I’ll accomplish that quite yet.

And this kind of wraps up what I’m thinking about for our AI Agents. I’ve been building them out with Copilot Chat’s help in the Copilot Studio. I started with just Huginn, asking some questions and then making an opportunity. Now I’m building out specific Workflows that will be shared by agents to do things like look up accounts and find contacts. This will help to keep them from creeping around where they shouldn’t be. Security here is very important. That I think is my biggest concern. While I’m working in Microsoft’s ecosystem, I’m not worried about training on my data or anything like that. I am worried about breaking these tools out of their boxes and getting to things they shouldn’t be able to get to. That’s why an “Access Everything” approach is probably not the best idea. Keep the agents in their confined space, don’t let them go rogue.

Moving forward, I don’t see Nordic branching into Agent development any time soon. I think that is going to stay with the application and web developers. Our role will be to help evaluate the security and accessibility concerns of the tools, and to help get them hooked up and keep them working. I do see how building in Copilot Studio has some great advantages though. Authentication is baked right in, security is easy, and there is basically no code to write. I’ll be really interested to see how well these tools work, and how much time they save us.

-Nate