Agentic AI Demo - Multi-Agent Orchestration in Servicely

In this video, we demonstrate Servicely's Multi-Agent Orchestration within the platform's Agentic AI model.

‍

Video Transcript

This space is moving at a pretty solid pace. As I mentioned, we were talking about generative AI—it was all the rage, as it were, literally a few months ago. Now we’re on to agentic. But I think what we’re seeing is it’s the coming together of all these technologies as a real enabler.

And for those more technical on the call, this just hasn’t happened by chance. It’s really been an evolution of the AI technology. It’s the coming together—when we look at agentic, it’s large language models, i.e., generative AI capability, with retrieval-augmented generation, which is another mechanism that’s become an underpinning capability of these systems, and then this notion of vector data stores—so really turning the data into embeddings and bringing this all together in a cohesive manner to really deliver business value to organizations.

And what does that look like? We’ve just set it up in this demo. This could be through a virtual agent, it could be coming in via an email, it could be whatever we chose, but this was a good way for us to demo.

So imagine if I said, “Create a new user for Dion Jones, and give him an email, and order him a security fob.” This is where we now start to be able to use the LLMs in generative AI, where we can now go through the process. And I’ll just assign that to support—we’ll just call that a request.

And what that allows us to do now is—whether we’re surfacing this via a virtual agent interface of “Hey, I need this, this, this,” or whether it be an email that’s coming into the system and the system—and the LLM—is processing that email and saying, “Okay, I know what to do with it,” or whether it be in a co-pilot type fashion—what we’re able to do is we’re able to use our agentic AI agent in this context to actually go and start to do some work for us.

And what we can see over here is the agent actually working in the background. And what this agent’s done is, first of all, it said a primary task has been processed by the multi-agent task orchestrator. And the way these agentic systems work is, you have kind of the main agent, which takes the information you give it, and it basically looks at that information and says, “Okay, what do I need to do with this information?”

And it has agents that it can control or can talk to, that are essentially capable of executing a task. So if we look over here, it took what we gave it, broke it down, and said, “Okay, create a new user for Dion Jones.” And it has an agent, which is the user provisioning agent, so it said, “I know you know what to do with this, so I’m going to give this task to you,” as the multi-agent task orchestrator or the main agent.

And then it said, “Well, we need to give Dion Jones an email.” And I see I have an agent that knows how to update a user’s email, so it’s passed that off and said to that agent, “Hey, you need to go and create an email for this user that’s just been created.”

And then it said, “But hold on, you’ve asked for something additional—and that is to order a security fob. I don’t have an agent defined that knows how to do that, so what I’m going to do is I’m simply going to go and I’m going to create a subtask, which is: order a security fob for Dion Jones. And that’s going to get assigned to a queue, and someone’s going to go and pick it up.”

And what we’ve just done is we’ve dynamically taken some English instructions and gone off and actually executed that.

And just very quickly, to give you some insight—effectively, the way this is working in the backend is: we have these agents that are sitting there. And many customers have—actually, I might use the “Report a Problem” one if I can see where it is, give me one sec.

You know, many customers often say, “Well, how do I do automated problem diagnostics?” Well, it just so turns out—why not have an agent that’s monitoring all the incidents that are being created, looking for similar incidents that are occurring over a period of time, and on the back of that, actually going and creating a problem to assign to the problem manager to go and do an investigation—and then linking all the records that it found, or the incidents, to this new problem, with a detailed analysis of what’s happening?

So this is where we’re now really getting the agents to start working for us.

And we have a similar agent in this system that we do for major incidents, where we say, “Monitor—if you see a certain number of similar incidents in a specific period of time, go ahead and create us a major incident.”

And if I had to go over here, we can see it had created—so this was automatically created—this incident: Major Incident – VPN Connectivity Issue, detecting affected four users with four reports in the last 10 minutes. The issue has been raised four times in the last hour, and we can see there’s a bunch of incidents that have been linked to that major incident, automatically created.