(So you can skip ahead, if you want.)

00:00 Introduction to FlowAI Demo
00:50 Creating Agent Configuration
02:10 Attaching MCP Tools
03:45 Agent Provider Setup
04:30 Slack and Email Notifications
06:00 On-Demand Agent Execution 
08:15 Wrap Up & Industry Impact 

Hi. John Capobianco here. And guess what?
I’m gonna build my first FlowAI agent in the Itential platform. I’ve been here three weeks. One week was spent in Atlanta. You saw me having the fun and William giving me the tour of the office.
And now ten minutes of overview and my first agent in a few clicks. Natural language. Now I’m gonna show you the video and I want you to stay to the end because I have a lot to say about what you’re going to see here. Okay? So watch the demo and stay tuned for my thoughts and my analysis analysis and really why this is an important step in our industry. Thank you.
Hi. John Capobianco here, and I’m just a few days and weeks into my time at Itential, and they’ve given me access to the preview of FlowAI, and we’re gonna go into the FlowAI manager here, and I’m gonna build an agent. Alright. So this agent’s name is gonna be called John’s First Agent, and we’re gonna add an agent personality and context and some agent instructions, right, how the agent behaves and what the agent should do.
Now I’ve pre canned a couple of these. The system prompt, I’m just gonna steal this from my notepad here and show you how easy this is. I could free type this, but just for the sake of the demo, I’m gonna type that in. And I’m giving it some CCIE level network automation agent specialization using network device analysis for the Genie.
And its primary responsibilities, when analyzing interfaces, what to consider, some summary details, and to use PI ETS. Now I wrote the PI ETS MCP, and we’ve easily, easily added it to FlowAI as an available MCP. And not just an available MCP, but we’re also gonna tell this to Slack the results and email the results. And you’ll see in a second how this is done.
Slack the results, email the results.
Alright. So I’m gonna go to the next screen in the builder and select my tools. Now I can select tools, which are tools that are from MCP servers. I can attach to deterministic workflows. I can attach to projects.
So I’m gonna attach this agent tool belt, and I’m also going to attach some PyATS tools, including the show run command, the show running config, and the dynamic tests.
So PyATS and the agent can write and execute PyATS tests dynamically given the workflow and the deterministic set of instructions.
Next, I’m gonna pick my provider, and this is all pre canned with Claude Sonnet. Alright. And we’re gonna hit next and review our agent details and create the agent. So if I go to my agent missions, we can see if I go to John here, maybe I didn’t actually launch the agent.
Hang on a second. Let me go to that agent and turn it on. So we’re gonna hit click starting John’s first agent, agent missions, and there is John’s first agent in progress less than a minute ago. I’m gonna let this agent run and cook, and I’ll be back with some results very soon.
Thank you.
And we use the, you know, some tokens, eleven tool calls with twenty messages. It took less than four hundred seconds.
And we can see here, right, the the system messages, the user messages, that it’s calling the PyETS show run commands, all the way down to that the notifications have been sent successfully, and check this out. There’s my notification in Slack.
Right? The full report here, and I should have an email waiting for me here as well.
So we have the key findings, we have the critical findings that this IPSec tunnel is down, and the healthy components, the healthy metrics, the error analysis, one interface with errors, error free interfaces. The next steps, immediate, investigate the tunnel’s IPsec connectivity issues, right, monitor the error trends, and continue to check this. The email’s been sent, the Slack notification’s sent, and we have a verdict that forty one out of forty one interfaces are fully operational, one medium priority issue, the tunnel zero IPSec tunnel, requires attention. All other interfaces are error free and stable.
Right? Now I can run this on demand anytime I want, and I have the full awesome reports and Slack notifications and emails.
Pretty awesome, right? Pretty awesome.
Email, and let’s take a look at the email that I got. So I got a Slack message, I got the email, here’s the full report, ninety five out of a hundred score, the metrics breakdown, the critical finding, a complete breakdown, all of the tests that it performed, everything that I’d wanna know, right? Look at this detailed report.
Unbelievable. Unbelievable. All from my new digital coworker, all from my new digital coworker, John’s agent. Right? So if I manage my agents and I look for John’s first agent, I can just fire this.
I wanna test interfaces, I just fire this. My new digital coworker will do the thing, in this case, run PyTS, generate dynamic testing, execute the testing, send me a Slack and send me an email. From what? I just wanna remind you, from what?
From a system prompt and a user prompt. Snapping in the right tools and the project agent tool belt. That’s it. That’s all.
That’s as easy as it is. I’ve been here less than a month and in five minutes, I’m building agents.
This could be you. This will be you. How cool was that? How cool was that? A little wizard with a natural language prompt for the system, a natural language prompt for the user. We attach things like deterministic workflows, projects, MCPs, but not just all the tools from the MCP. We surgically select the tools we need and bind them to the agent.
I click run, less than four hundred seconds go by, and I’ve just tested every interface with dynamic testing.
The agent wrote the tests. I didn’t write did you see me write any tests? The agent wrote the tests. The agent executes the test with real data from network that PyATS has parsed.
Now, as someone who literally wrote the book on PyATS, this is incredible. A couple of things. One, I didn’t need to write a PyATS job. Two, I bound a testbed YAML file to this agent in the back end.
So it could be one device, it could be every device in your network. You could test every interface on every device through my new digital coworker, John’s first agent. Now I asked the audience at AutoCon, who here knows the health of every interface on their network? Not a single person raised their hand because it’s hard.
That to do what I just showed you with PYETS. And I’ve done it with PYETS. Look at my catalog. Look at my history of my work.
Look at my videos.
Have pyATS test things. Send us Slack. Send an email. That is hundreds of lines of Python.
Did I use a no code, low code? No. Did I write a single line of Python? No.
Did I write an Ansible playbook? No. Did I use any REST APIs? No. I used my natural language as a system and user instructions.
I bound the tools, which are easy. You saw me bind the tools, and I hit run.
Think about it. Think about what this means. I have a new digital coworker that if there’s a problem, someone says, could you check the health of the interfaces? I’m not digging up the device name.
I’m not SSH ing in. I’m not running show commands and reading parsing them by hand, copying them out of the terminal into a notepad, and then analyzing them. I’m not even doing it the old the new way with an external call through MCP. I bind the MCP to the agent and give the agent its guardrails and its instructions.
This should really open your eyes as to why I joined Itential in the first place and my previous messages about how excited I am about FlowAI. So this was my first agent. Just wait till I cook some more agents up. So follow me. Follow Itential. Follow Ankit, Jaxon, William, Peter, Chris Wade, Kristen Rachel’s, the whole team at Itential because we are changing the way that infrastructure is operated and monitored and tested at scale through agents, new digital coworkers.
Thank you.