Put FlowAgents to Work on F5 VIPs: Faster Changes, Less Risk

John Capobianco • 00:00

Hey everyone, I’m John Capobianco, head of AI and DevRel, and here with Joksan Flores, our principal solution engineer. We are doing a quick vibe today. How many of you have an F5 on your list of responsibilities, right? And we’ve got VIPs and SNATs and reverse snats and all this great stuff. What if we could move this into the Agentic space? F5 has wonderful APIs. We have integrations with our platform.

John Capobianco • 00:25

They have an MCP capability as well. Joxon’s going to walk you through some VIP management, lifecycle management through our platform and through using FlowAI agents. Now, you’re also going to see our work center today. That is our human-in-the-loop capability. So human-in-the-loop is definitely an option with the Flow AI agents where you want to interact with them. You want to give them certain information. You want to approve or deny their actions.

John Capobianco • 00:51

You maybe want to give them some more guidance or see their feedback. So in just a few minutes, we’re going to do some VIP management on F5 through a FlowAgent.

Joksan Flores • 01:06

So we got a VIP management, a human-in-the-loop agent. So essentially, this whole idea here is that this will do an entire lifecycle of the VIPs, right? So it’ll actually create VIPs, delete VIPs, manage them, and it’s actually going to the load balancer as a source of truth. So I’m actually going to go ahead and click run so that this thing gets started. And you’ll see that there’s a lot of stuff that’s going on, right? It’ll do pre-flight checks. It’ll do the creation logic.

Joksan Flores • 01:33

It’ll do deletion and so on and so forth. So it’s got the capability of doing essentially everything as an Agentic operation while also leveraging the deterministic assets we have in the platform. So what I’m going to do here, I like to do this way, John. So I’d like to split screen. And then I have my work center on the side, and I have my agent execution because I like to serve my agent. So, if you look at the agent, John actually went through a bunch of reasoning stages, pulled data from its knowledge source, which is a repo here, and it’s got inventory from the platform as well. So, it did its holds, its data calls for the initial pre-checks, and it’s actually asking me to go to complete a task in Work Center.

Joksan Flores • 02:09

And so, let’s go and see what that looks like. And we have it right here, John. So look at that. We got a form that gets rendered dynamically from the agent itself asking for more information. So, what I’m going to do here is I’m going to select my lab device, which I think we only have one.

John Capobianco • 02:24

So, you didn’t ginger template this out or anything like that at all, right? It’s all dynamically driven by the agent.

Joksan Flores • 02:30

Yeah, these are dynamically driven. Of course, you can use, we have deterministic all throughout the entire platform. So, you can ginger templatize them if you wanted to, but these are not. These are just, hey, we told the agent, this is the data you need to get. Go pick the best way to do that. So, let’s go ahead and we’re going to create a VIP. It’s what we’re going to do:

Joksan Flores • 02:44

F5AWS1. Let’s go ahead and complete that. And you’ll notice on the left side that the agent will continue reasoning. So, John, this is the human-in-the-loop interface, which is entirely separate from the agent’s session management itself.

John Capobianco • 02:57

Yeah, I love this split screen, by the way. I think I’m going to drive this way. I’ve been bouncing back between the two tabs, but this is nice.

Joksan Flores • 03:05

Yeah, I like to show the two personas, and I always you have to tell people, you have to frame it, right? When I talk to customers and stuff like that, I tell them, Hey, this is what an admin or builder would look like when they’re building. But a consumer themselves are going to live probably here, right? They’re going to go with, hey, I’m going to request a VIP. I’m going to wait for my service to do things and/or workflow, right? The work center actually works for anything, workflows or agents, and they’ll live here making decisions where in the background, this is what you’ll see. This is what will happen, right?

Joksan Flores • 03:33

For auditing purposes and so forth. So, you can actually scan all tool calls and so forth back here, right? You can see everything. Yeah. Let it do its thing.

John Capobianco • 03:46

There we go.

Joksan Flores • 03:49

Yep. So now we get a provision and request to John. So it’s actually getting full-on parameters for everything because we said we’re going to create a new VIP. So we need to provide everything that we need. So we’re going to do a sales VIP in dev. We’re going to do our VIP is going to be 10.10.20.1. We’re going to do a manual assignment for the IP because otherwise it’s sent us to an IPAM.

Joksan Flores • 04:09

And we could do that as well, right? We have the capability. 443 HTTPS. We’re going to do, and then we can do a table of pool, right? That makes sense, right? We need members in the pool. We need more than one.

Joksan Flores • 04:20

And we’re going to do colon 8000. And we’re going to do, oh, the port is not here. It’s on here. And then we’re going to do another member, 102.168.51.3 and port 8000 as well. The health monitor is going to stay HTTPS. And then my host name is going to be sales.se.itensol.com. And my DNS zone is still default.

Joksan Flores • 04:45

I’m going to do a record and TTL of 300 seconds. And I’m going to make up a ticket here because we’re not doing that right now. But look at this, John, here, where we actually the agent itself is telling us, hey, this is mandatory for all production environment changes. So you have to put a ticket on that has been approved. Obviously, we could do the ticketing through the agent itself, but in this case, we wanted to demonstrate a separate motion, right? Because some customers necessarily want to automate the ticket approvals. So complete all that information and let’s let it go through.

Joksan Flores • 05:21

Let’s see. Eventually, it is supposed to execute a check for that sales VIP and verify whether it exists or not. It should not exist at this point, but maybe somebody was doing some testing or playing around and found it. Let’s see. We’re using the S3 module. Yep, the VIP 10th sales does not exist. Confirm 404 not found.

Joksan Flores • 05:45

Okay, that’s great. So it’s safe to proceed. This is a great thing about reasoning, John. You know, we had to do these things with evals and things like that. So, oh my gosh, we got to eval and query our way through all these things, right? So, combining reasoning with some of these tools, right? Like, we have AS3 direct access into the lobancer with control-deterministic tools.

Joksan Flores • 06:04

Okay, look at that. I created the VIP.

John Capobianco • 06:07

Created it. There we go.

Joksan Flores • 06:10

That’s doing a DNS. Oh, it’s actually calling DNS agent. So, it’s actually calling DNS A record agent to go and provision that DNS for it, John. So, it did everything. Yeah, the whole motion. Yeah. Really cool.

John Capobianco • 06:23

And we’re at 12 tool calls, 6,000 tokens, right? And then, and if you’ve watched our previous webinar, we show you how this could be done completely for free with various open weight models. We just had a really good webinar about that. So, this could be, this could cost you nothing to run. And look, it’s done. It’s pretty much done. We’re going to get the message with the data and a nice report.

John Capobianco • 06:44

It would have attached all this to the ticket and closed off the ticket should you want it to. And it even gives you the testing information, right? So, the workflow summary, everything succeeded.

Joksan Flores • 06:56

Yeah, and you could actually go back and see, John, the other day I had a customer talk about DNS. I’m going to act this so that it continues. The other day, a customer was asking me about doing synthetic testing from the internet once they created DNS and things like that. So it’s like, hey, you can add all those things, right? If you want to, if you have an endpoint somewhere, you can add synthetic tests into it once you’re provisioning DNS for it. So imagine if you’re publishing web services that are going to be exposed to the internet. You could do that.

John Capobianco • 07:21

And this took 6 min , everyone. We estimated 10 minutes for this video. It took 6 min . It feels longer because some of them are waiting sometimes while it reasons, but that’s a good thing. So 8,200 tokens, 13 tool calls, 6 min . We provisioned a VIP. When was the last time you did that in 6 min ?

John Capobianco • 07:40

We tried to automate as much as we could with the F5. It actually does have a nice REST API system, and we did do some pretty good automation. But that automation took me a day to write and it had to be maintained in every new edge case, every new error. Right? Unbelievable that we just do this egentically now. So, thanks for joining us for this quick vibe. This was about VIP lifecycle management using Flow AI agents with human in the loop.

John Capobianco • 08:05

Joxon, thanks for driving. It was really good to see you again. Thank you, John.