Inside Work Center: Human-in-the-Loop Control for FlowAgents & Workflows

Dan Sullivan • 00:02

Hi, everyone. Welcome. My name is Dan Sullivan. I’m part of the solutions engineering team here at Itential, and today I’m here to speak a little bit about a new platform feature called WorkCenter. The most recent platform release on July 1st was really focused on our agentic orchestration offering, Flow AI. However, we also released Work Center at that time, which provides a vital role for both agentic and deterministic orchestration within the Itentral platform. The Work Center interface provides human-in-the-loop support for both deterministic orchestration and agentic assets within the platform.

Dan Sullivan • 00:42

It’s a unified method for basically manual task execution, interacting with agents or workflows or both. If you’re at all familiar with the previous versions of the platform, you know that you would typically work manual tasks from Operations Manager, which is really workflow-centric. So Work Center actually provides that unified view for both types of assets within the platform without that workflow-centric view. So we can actually jump in and take a quick look at it. So if I jump into Work Center here, in this case you’ll see that there’s really only one manual task that requires any input. And that’s this one here. I won’t necessarily work that this 2nd , but what you can actually see is we’ve got a number of filterable fields.

Dan Sullivan • 01:37

You can filter all the available manual tasks outstanding. those that are assigned to me and those that are unassigned. On each column, we have the source, the specific initiator, what type of trigger it was, the person who ran it, if it was manually started, the origin, which is the root job or agent session that the task belongs to, the elapsed time, how long the job has been running, the assigned to field that you can certainly change, and then how long it’s been in queue. So how long it’s been waiting for some input, and then the actual action that’s waiting. You can also filter on all of these fields and sort that sort of thing. And when you actually open one of these, so if you were to click on this, you’ll get some information about the job at the top. And in my case here, there’s a form that needs to be filled out.

Dan Sullivan • 02:42

And if I fill it out and submit it, it will go back and continue execution. Before we do some of that, what we can 1st do maybe is take a quick look at a project I have that I’m using to trigger some of these manual, some of these operations. So let’s take a quick look here. So we’ll just go to the projects home here. And I’m playing around with this Python, this Python one. So I just have a little bit of a workflow that isn’t going to take any user data. It’s going to actually prompt.

Dan Sullivan • 03:27

And that’s that form that you saw when we were looking at Work Center. And then as I make some, run through some evaluation code here, I’m actually going to have some output, and then potentially I’ll either exit or provide the user the opportunity to go back and run it. So that’s pretty much what we’re doing here. So if I were to start this, we should see a 2nd enter and a 2nd entry in Work Center. So if I go back to Work Center here, click on the top level, you’ll see, oh, again, we have a new instance of the same workflow. So in this case, if I want to go here and collect on the user input, that form will show up here. I can type some data in.

Dan Sullivan • 04:06

My workflow pretty much just checks for overlapping subnets. So nothing too fancy here. I’ll just type in some random stuff. Let’s see what happens. Let’s see. I’ll put in something I know it matches with. Maybe a couple of other examples.

Dan Sullivan • 04:25

Let’s type quickly here. Shouldn’t overlap with this. And now I’ve got some data. If I hit submit, the workflow is going to keep running. Now, as I told you, this particular workflow has another step in it. And you can see here that I’ve kind of got a pretty nice user interface so I can do multi-stage operations. The other benefit here is that I can collect information using the workflow, using integrations, and then provide that as data in a form.

Dan Sullivan • 04:57

In my case, here it looks like it’s found that overlap, that one prefix, and it says that the other prefix doesn’t match. And if I want to, if I want to retry it, it should go back and I should see more collect user input here. So let’s see what happens. So if I hit that in just a 2nd , you’ll see that same form. Now I had it so that it plugged in the same data I had, but I might want to change this. Maybe I want to add something else. 16.

Dan Sullivan • 05:37

See what happens here. So now I’ve got, I’ve added a new prefix and I’ll run again. And I should see the, I should see those results again. So this is the active task. You’ll see the green is things that have been completed. I can go back and look at the other data that I’ve entered or the results pretty easily here. So that’s kind of nice that I can go back and look and see what I’ve done here.

Dan Sullivan • 06:01

And now at this point, I think I’m good. And I can actually just complete it here. That’s kind of what happens when you start building out workflows with multiple manual tasks. You can start to sort of almost create a wizard-like experience here. And it will keep going as long as you have manual tasks. Now, the same functionality is available from agents. And so I have a little bit of an agent example maybe we can take a look at as well.

Dan Sullivan • 06:29

So let’s just jump into Agent projects here. I’ve got kind of this firewall provisioning agent that I’ve been working on. And this particular agent, if you look at the prompt here, I don’t want to get too deep into agents, but there’s a bunch of data that it needs. Among them, the source organization, the destination organization, and even the application type that I might want to filter. But the problem is that this is a fairly substantial form. There’s a lot of data to type out, and especially the organizations where I have a list of those kept in Git.

Dan Sullivan • 07:09

So in this case, I’m actually telling it, I’m telling the agent, if you don’t get the information you need, you should go and prompt the user to enter it. So up here, you’ll see where I’ve basically just told it, hey, you need to go and get the information offer and provide a form if the data isn’t entered. So you’ll see here in the prompt, all the configuration data lives in GitLab and must be fetched at runtime. So you don’t really want to assume any of that. So if I just run this agent, it doesn’t really, I didn’t give it any data. So it’s quickly going to figure out that it needs to prompt the user for that. So it’s actually reading some context from GitLab.

Dan Sullivan • 08:06

And you’ll see here, you’ll see the next step here. We should see it come up with a task for Work Center. So it’ll actually create a form and prompt me to go and fill out the data in Work Center. So now I actually have an Agentic agent, and I’m able to prompt data from it. And it’s going to build out a pretty interesting form, and I’ll explain that to you in just a 2nd here. And so that’s using some of the same manual tasks that you might be familiar with in In a more normal workflow, a more traditional workflow.

Dan Sullivan • 08:45

But some of those same tools are available. So now it says it’s rendering JSON schema. So it’s effectively gone and grabbed some data and has created a form for me. I can go over to Work Center here. And again, you’ll see the firewall provisioning app here. And let’s see what it comes up with. So it’s telling me, hey, this is the data you need to add.

Dan Sullivan • 09:08

So let me put in here source organization. So it’s actually fetched this organization information, though it spelt it wrong, from Git, and then created some drop-downs so that I don’t have to type them in, and I won’t make a mistake here. So let’s see, maybe we’re going to try to get access to the I’ll just quickly fill out some so I’ve kind of filled out the justification for this firewall and then what it’s going to do it’s going to run for a little bit and we should actually see another manual task here. So the and we’ll take a look at that. So I’ve gone off here and given the agent some data if I wanted to go back to agent sessions you’ll see that it says hey the tool call is Was successful.

Dan Sullivan • 10:28

If I look at this, I can see that the data that I passed in from the form, so I was actually able to prompt the user, create a have the agent dynamically create a form, prompt the user, get the data back. And so it works out as a pretty good way to be able to interact with agents now. So if I collapse this a bit, let’s say the agent’s kind of running off. It’s looking like it went off to get the pre-rules and figured out that they don’t exist. So that’s not a bad thing that the tool calls fail because it’s still going to try to provision something. And once it figures out what it wants to do, it’s going to come back. If we were to look at the prompt again while it’s running here, it’s going to come back and actually look for some, it’s going to prompt the user to actually approve what’s going on.

Dan Sullivan • 11:21

So down in this step here, it’s doing a bunch of stuff with ServiceNow , calculating what it’s going to do, and then it’s going to stop and await some approval. And it’s going to use a manual task to do that. So let’s see if we go back to agent sessions where we are. So here’s what’s running. It’s almost there, I think. Takes a little bit of time, but maybe hopefully the juice will be worth the squeeze on this in just a 2nd . So it looks like we updated the change request.

Dan Sullivan • 12:00

So it looks like, yep, so we’re in pretty good shape here. I think we figured out what we’re going to do, and now we’re in pending approval. So I can go back to Work Center yet again. And now I’ve got some information about this particular firewall change that I’m going to. I can decide if I want to approve this or not. It’s got the actual information that it’s going to push to Panorama in this case. And I could proceed with this if I wanted, or I can just deny it.

Dan Sullivan • 12:33

I think I’ll just deny it in this case. And that should be it. This gives you a little bit of an idea about how you can now interact with both agents and more deterministic assets like workflows. We even have some new tasks that are available to help you generate user. To give you reasonable views of user data and things like that. So, and maybe in some subsequent videos, we’ll dig in a little bit more on that. But I just want to give you a quick overview of.

Dan Sullivan • 13:06

Work center and the types of things that you can do with it, and sort of the experience that you might expect. Thanks, and we’ll see you soon.