Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
The Joy of Prompting: Network Automation and AI Agents with John Capobianco
8 Plays15 days ago
Everyone is racing to teach AI to write code. Almost no one is teaching it to talk to the network. That blind spot the infrastructure the code actually runs on — is where the next decade of automation gets won or lost.
In Episode 4 of The Root Cause, Priyank Upadhyay sits down with John Capobianco, Head of AI & DevRel at Itential, longtime network automation evangelist, author of "Automate Your Network" and the Cisco Press pyATS book, and the engineer who spent 25 years going from analog modems to agentic infrastructure. John was early to network programmability, early to AI in networking, and is now building agents that let you literally phone your infrastructure and ask it what's wrong.
We get into the Bob Ross school of automation (leave the happy little accidents in, paint around them), why prompting is the great equalizer that lets engineers skip years of Python and DSLs, how he built NetClaw to talk to networks over Twilio, the difference between giving an agent skills versus scripts, why "just start" beats waiting until you're an expert, and his relentless optimism that this is simply the next era — coax to CAT5 to wireless to SDN to agents.
The Root Cause is an original series by RubixKube — Site Reliability Intelligence: see more, plan better, act safely, and learn with every incident.
Platform: https://rubixkube.ai
Community (Slack): https://rubixkube.ai (see footer)
#AIOps #SRE #SiteReliabilityIntelligence #DevOps #Kubernetes #NetworkAutomation #JohnCapobianco #CloudNative #AIagents
Recommended
Transcript
Technological Progress and Personal Impact
00:00:00
Speaker
It took you an hour, an hour and a half to figure it out. And then you had it working and you had your network phoning you or your other MCP system actually phoning you. And, you know, and ah ah that that really excites me. So this is going to be our worst attempt at it because it's our first attempt at it, right?
00:00:17
Speaker
So I think it's only going to get better. And I think it's only going to get cheaper and faster and more available and and more quality answers over time. Technology can...
00:00:28
Speaker
can can break the chains of poverty. that can that can You can use your mind instead of your body, right? um you know I was almost killed in that place. I can't tell you how many how many dangerous situations that you get into in ah in an aluminum factory like that.
00:00:44
Speaker
A CD-ROM has never come shooting out at me trying to kill me, right? like a That's something very abnormal to me because somebody who has spent 25 years in pure backend, pure networking, i i typically see dead souls.
00:01:14
Speaker
Awesome. Sounds good.
Introduction to John Capobianco and His Journey
00:01:16
Speaker
So welcome everyone again to the fourth episode of The Root Cause. And we have a very interesting guest today in the podcast.
00:01:28
Speaker
I'm very excited. I'm truly very excited to meet him. I had one call with him last night and it went very well. And today i have him on the show to answer questions. I want to know him more, how he thinks about things.
00:01:47
Speaker
And yeah, let's let's start it. Well, thank you so much, Priyank. And you've been so kind. And um it really excites me. We were just talking before we started about how you saw something that I had done with Twilio.
00:02:03
Speaker
It took you an hour, an hour and a half to figure it out. And then you had it working and you had your network phoning you or your other MCP system actually phoning you. And you know and ah that that really excites me. So ah my name is John Capobianco. I'm the head of AI and DevRel for iTential.
00:02:20
Speaker
I've been in in IT for about 25 years now, and this is actually my twenty celebrating my 25th year and actually working, for getting paid for i work in IT.
00:02:32
Speaker
And um for the past, let's call it three years, almost four years, it'll be four years in December, I've shifted my focus uh, not away from networks or network automation, but more towards artificial intelligence and seeing where the two might intersect, where this new technology and the new tool sets that have come out. And there's been a lot of them that have come out in the past three or four years, how they can apply to my field and my trade of network automation.
00:03:02
Speaker
Um, I, I've happened to share this journey publicly. I've recorded every step of the way. I've made, um, many GitHub repositories and posts and videos, because I think it's important. I think we need to realize that we're literally in a new era. And this is at many layers. It's not just our industry that's going through this or just network engineers that are going through this. It's everyone that's going through this.
00:03:29
Speaker
Society is going through this, multinational corporations, governments, healthcare industries, manufacturing, agriculture. Everywhere wants to embrace the power of artificial intelligence.
00:03:42
Speaker
And if we boil that down to what we do or what I do, um I think it's a very exciting time, really exciting time. And we should be enthusiastic and we should be open-minded and we should be flexible.
00:03:57
Speaker
and and and we have But we've always adapted. my 25 years, Priyank, I can't tell you the number of changes that the network industry has gone through. And I don't understand why some people don't see this as just another one of those changes.
00:04:11
Speaker
Like we went from thick coax cables, right? That we put vampire teeth into bus systems to cat category 5 tables and cat 6 cables and twisted pair and then wireless came along and then voice over IP came along and then spine and leaf in the data center came along and software defined networking came along with controllers.
Passion for Technology and Career Reflections
00:04:34
Speaker
there's there This is just, that's all this is, is another new capability and technology um that I think we should we should really embrace and and take advantage of.
00:04:48
Speaker
I totally agree, John. And first of all, am so glad you're here and welcome to the root cause. Thank you for making it out today.
00:04:59
Speaker
I know you're a busy person, John. There's one other thing that excites me apart from your portfolio, your resume. Yesterday, i saw the spark in your eyes when you were describing the why do you love doing tech, right?
00:05:18
Speaker
And that's something very abnormal to me because somebody who has spent 25 years in pure backend, pure networking, I i typically see dead souls.
00:05:33
Speaker
So... Well, i thank you. i um you know i don't think it's unique to me. I think that I happen to be very animated, lively, excited personality. But this isn't a show or an act. It really, truly, this means something to me, connecting with you in India right now, across the world, and sharing our ideas and having this conversation about where we are in the world.
00:06:02
Speaker
um But yes, some i've I've experienced that. I see some people that have maybe these lost lost the spark or lost the connection to what excited them about technology in the first place or networks in the first place.
00:06:17
Speaker
um I don't know that it's prevalent or that it's a lot of people, but yeah, certainly some people are ah you know jaded or they've done this a long time or maybe they didn't make the money that they thought they would make or they didn't have the influence they thought they would have or they look back and it's all ephemeral. You know, those networks they built have been replaced.
00:06:37
Speaker
They get replaced and and new people come along and, you know, what is your legacy? What are you leaving behind? there There's a lot that goes, that that there's a lot of um emotion connected to our our job, right? and And what we do for a living and and pride. and And I hope there's some pride involved.
00:06:53
Speaker
um I happen to be fortunate that You know, I, I did work very hard, um, and got to work in some very exciting ah environments. You know, I got to work in the, in the realm of, uh, insurance and finance, uh, running a network there for about five years. And then, um, you know, the parliament for, for eight or nine years and then some time at Cisco and then at some time at selector and now I potential.
00:07:19
Speaker
So I, I've been very fortunate and and I've worked very hard to, to get these opportunities and, um, It's hard not to be excited for me. I, cause I can remember when I was a young man, you know, like I can remember what it took for me to get here and things I've gone through in my life and and endured in my life to get to this opportunity where I get to build AI agents. I'm working with artificial intelligence. Like that, that should make you excited and happy.
00:07:47
Speaker
is It's so remarkable. The, the models that come out and the new frameworks that come out and, and the human ingenuity that, that we figured out ways to augment these AIs and, and,
00:07:59
Speaker
attach these AIs to network infrastructure now, right? I think we should be excited. And the other thing is, I don't think that my message would resonate as well if I didn't express it with with the emotion that I do.
00:08:14
Speaker
I really want to connect with people. I'd like to people to perk up and pay attention and listen. And and prove that through the camera, and this is something that I learned actually just recently recording with Google, um you lose about 10% of your energy through the camera.
00:08:31
Speaker
Like when people are watching this back, you know, there's something lost in translation versus human kind emotion and connection. So you do have to turn it up a bit. if if you want to be paid If people want to pay attention to your YouTube content,
00:08:46
Speaker
You know, if I'm not excited about it, why the hell would anyone want to listen to it or pay attention to it, right? Like, you know what i mean? Like, if the person expressing the ideas isn't at least excited or enthusiastic about it, then why would people pay attention in the first place, right?
00:09:02
Speaker
Absolutely. Yes, I totally agree to that. John, you mentioned your career journey. You worked at Parliament, you worked at Cisco, you worked...
00:09:13
Speaker
a lot of places and it's very exciting. Yesterday you mentioned something very weird. You started your career with in an aluminum factory.
00:09:23
Speaker
How is that? How did that come to be? Well, yeah, so i um it gets I was a very young man and um I found myself on academic probation from university in Ontario. So I went to Queen's University for a year.
00:09:41
Speaker
And ran afoul there a little bit and ended up on on probation from school. And then, you know, i ah I was working retail in a mall making, you know, minimum wage.
00:09:53
Speaker
And I was able to get into the aluminum factory and in my hometown. and get into the production line and and the union there. So I really needed that because I had to pay off my student loan and and get an apartment and all that other stuff.
00:10:09
Speaker
So I you know i i worked 12-hour shifts, and I can remember the rotation because it was Monday, Tuesday, day shifts, so 7 a.m. m to 7 p.m., Monday and Tuesday. And then you'd have Wednesday, Thursday off, and then you would work Friday, Saturday, Sunday, midnights.
00:10:27
Speaker
So 7 p.m. to 7 a.m. m for three nights in a row. Then you get Monday, Tuesday off and work day shifts Wednesday, Thursday, have the weekend off, and then switch the rotation again.
00:10:43
Speaker
And that was it. You'd go in there for 12 hours, and you'd help run aluminum. And it was dangerous work. We had gloves and steel-toed boots, sometimes helmets. I drove forklift for a while.
00:10:53
Speaker
I worked on what was called a tension leveler. So that was my job was working on machines like that. And and after about two years of that, i really, you know, I was still, I was only 22 when I decided to go back to school.
00:11:07
Speaker
I worked there for five years and my last three years I went to community college. So I made arrangements with the union. and the company to give up my week of day shifts.
00:11:20
Speaker
So for three years, I would go to school from, you know, eight to four, and then I'd go to the factory from four to 12 or, or from 12 to eight at the factory and then eight to four at the school.
00:11:32
Speaker
So um I have a very serious appreciation. Yeah, well, i that when people ask me why I'm so excited about this is because I'd like people, other people, maybe outside of the industry to hear my story, that you can go back to school, that you can change the traje to trajectory of your life, that technology can Can can break the chains of poverty that can that can you can use your mind instead of your body. Right. um You know, I was almost killed in that place. I can't tell you how many how many dangerous situations that you get into in ah in an aluminum factory like that.
00:12:07
Speaker
But I, you know, a keyboard, a CD-ROM has never come shooting out at me, trying to kill me, right? Like a keyboard has never malfunctioned in a way that would take, you know, physically injure me,
AI and Network Automation Challenges
00:12:19
Speaker
right? So I take my, my you know, what i earned through that, that I earned the right to work in in a maybe a different class of job, to transition from physical labor in an aluminum factory to IT work and network engineering.
00:12:37
Speaker
I take that very seriously. That's very important to me. um You know, the the the diploma and and the skills I learned as a programmer. Now it's funny. i only took one semester of networking during that three years.
00:12:48
Speaker
All the other courses were related to anything from COBOL and KIX and JCL to, um it was VB6 at the time. know, had computers beforehand then, right?
00:13:00
Speaker
I grew up with computers. Absolutely. We, um, We had a bulletin board system when we were kids, my brother and I. We got a computer very early in life. We we were obsessed with the computer. um I can't tell you how many hours I've i've spent on a computer, honestly, like from from the age of eight. It's about 40 years of using and always doing cool things with computers, too. Like even at a young age, we built a bulletin board system. And i I don't know, that probably predates you. Did you ever dial into someone's computer?
00:13:29
Speaker
I did not. I started with direct cables from telephone lines. Okay. Yeah. So at early days before the internet, before the World Wide Web, you could install software and set up your computer with a dial-up modem. Yeah. And people could phone your computer with their modem.
00:13:51
Speaker
And they could play games on your computer. You could share files with them. You could interface with them and chat with them. So that was a ah precursor to the internet. And my brother and I had a ah very, very popular one.
00:14:03
Speaker
We actually had two phone lines at one point in time at the peak of the BBS craze. We had a so high speed line and we had a regular line and he had to be a subscriber for the high speed line.
00:14:14
Speaker
It was, a so yes, i I've always grown up with with computers and with technology. And um artificial intelligence to me, it just feels like I said, it feels like just another, the the latest innovation in this trajectory of compute and storage and networking, um you know, memory and compute and graphics cards and all these things have led us to this new innovation, right? And I don't, I guess I don't, I still don't understand the psychology behind anyone that still is not convinced.
00:14:50
Speaker
you know Do you have any insight into that, Priyankh? You just had discussion with someone that that didn't like the term VibeOps, for example. They had a they had a visceral response to just the terminology of VibeOps.
00:15:03
Speaker
So you've talked to people on both sides of the scale. What what what do you think is what it is? Yes, John. So just in my last podcast, I had Randy and Randy Bass, and it was a very interesting call, and he's a...
00:15:19
Speaker
Great man. he he I loved the talk with him. But interestingly, so I know you because ah first of all, i I thought last year I have coined the term Y-Bombs. I never knew that you already are maintaining a Slack channel and a community around it.
00:15:37
Speaker
So, um you know, this Y-Bombs is something where I and you also engaged on LinkedIn, right? And this is something which I presented on a panel, CNCF panel, ah just a couple of weeks back.
00:15:53
Speaker
And I got randy Randy's comment on my post, like he was very ah reluctant on adopting the term vibe ops. He's so afraid of nobody should be vibing in the operations area.
00:16:08
Speaker
But when I spoke to him, his mindset and approach to automation, to AI, is much ahead of what you know new folks are thinking about. He is full on, hands on, operations should be automated. This is not something humans should be doing.
00:16:27
Speaker
This is visit the task of a labor, should not be done by humans. So he is full on agreeing to the point at which the vibe ups resonates about.
00:16:42
Speaker
It's just that maybe he doesn't like the term or maybe he likes now. but yeah Well, i I did watch the episode and I have the utmost respect for Randy. you know he He coined some legendary phrases in the industry we still use and we still adhere to. And I completely agree with his sentiment that all of this stuff should be automated. It's been my message since probably 2015. I don't know when I started AutomateYourNetwork.ca, but um i've I've been very much banging the drum that networks need to be automated and that there's multiple ways to automate them.
00:17:20
Speaker
Now I think people can skip to the front of the line. And here's what I mean by that. The barrier why so many networks are still not automated is because people need to learn either a domain-specific language or Python, pure Python, and they're trying to maintain their network certifications. They're trying to learn about eVPNs and BGP and BXLAN and IPv6. There's already enough in the field of networking to learn and to master.
00:17:53
Speaker
Now we've asked these poor people, and I've asked some of these poor people, To also learn how to develop software or program or write scripts or write Python or Ansible or Terraform or use the cloud and use Python and REST APIs and Postman and Git and GitHub.
00:18:13
Speaker
Do you understand where I'm going with this? It's not in their classic traditional lanes of certification paths and trees, expanding their skills and being certified along the way.
00:18:27
Speaker
Yes, there is certifications for this automation stuff, but it takes anywhere between a year to three years to five years to, infant like I'm still learning Python. I still wouldn't consider myself a good Python programmer.
00:18:40
Speaker
um Even all the PyATS that I've written over time, right? And even though I've written two books about this and a lot of other people have contributed massively to this idea, the network automation forum, what what Chris Grundeman and Scott Robon are doing,
00:18:56
Speaker
a forum for network automation discussion, an actual event twice a year. You know, why don't we see adoption? So back to the point of skipping to the front of the line.
00:19:06
Speaker
Now I can use my it my my massive brain of networking knowledge and domain-specific powers and creativity and knowledge and ask an agent to write code for me.
00:19:21
Speaker
or ask an agent to ah help me automate something, or have an agent write that YAML file or that markdown file or that Python file. It's all abstracted now, right? And when you have agents in between, when I can just say to the agent, hey, agent, you know, in some complex network question, right?
00:19:41
Speaker
And the agent says, let me fetch the data. You know, I'll use MCPs and I'll use skills and I'll go get all the information so I'm not making things up. And then I'll answer your serious question with a serious answer.
00:19:51
Speaker
that's That's as as high a grade as the expert or novice asking the question. There's a spectrum here, right? Juniors might know a certain i amount to ask it, and experts are going to know that much more to ask it and interrogate it.
00:20:07
Speaker
So... We can jump to the front of the line. i think our new job as network engineers is managing agents. It's not managing infrastructure anymore. There's a layer between us now, right? For all the departments that we had earlier, because even in operations, your on-call, your incident management, your automations for DevOps, everything can be automated now.
00:20:36
Speaker
ah people are no longer required to put in that manual labor because these are very grunt work tasks. I personally feel infra- should be self-sufficient. It should come to you when it has a problem with a particular you know root cause and everything detected. This is the action human needs to take.
00:21:00
Speaker
I had this problem and that's what I tried to solve. So I think things are turning around, John. It's just that You know, you have spent 25 years doing the same thing over and over again, and it becomes a very you know difficult for people to come out of that mindset. I think that's one of the reasons people are, you know a bit reluctant on making this transition, making this jump.
00:21:27
Speaker
And that makes sense that the history involved in the, the, I don't want to say sunken cost fallacy, but, um, I am optimistic. I see more and more every day people making, uh, extraordinary projects that involve artificial intelligence.
Automation Protocols and Strategies
00:21:43
Speaker
Uh, everyone's making amazing dashboards.
00:21:46
Speaker
They're making amazing, uh, ways to ingest telemetry. Um, You know, I'm really excited about what's going on right now. We we talked earlier the in the preview about how I think MCP, that model context protocol, ah has led to a Cambrian period, an explosion in innovation, because now we can plug this simple USB-like key into a and then into an agent.
00:22:13
Speaker
And just ask the agent natural language questions. you know ah There's been an explosion in MCP and in just the past month, six weeks in my industry, you know network automation or networking.
00:22:26
Speaker
I don't want to name everyone because going to miss some, but but some major companies have started to introduce their own MCPs. Some companies have offered that maybe behind a certain, you have to have a certain level of licensing with them all the way to just here's our MCP is free use it, right?
00:22:44
Speaker
So model context protocol is very exciting. I think skills layered on top of it is even more exciting. And a lot of people ask me the difference, or do we still need MCP? Can't we just plug skills into APIs and things like that?
00:22:59
Speaker
i would I would suggest that MCPs are like the the keys on the piano and the piano itself. They're the tools, right? So the MCP server might be the piano and the individual tools are the keys.
00:23:11
Speaker
The skills is the ability for the LLM to read sheet music, right right? And that sheet music is going to be your prompt. I'd like to hear a song that has a lot of tempo with high notes and some bass and whatever.
00:23:25
Speaker
You know what i mean? Like the skill is, and here's and here's the here's the sheet music that I have. And MCP, here's how to play the, I'm going to play those tools, press those individual keys on the keyboard.
00:23:39
Speaker
I don't know if that resonates with anyone, but- I like the analogy, John. This is wonderful. I am a musician by the way. so So you sort of get it, right? Like the the guitar and the strings are the MCP and the tools. And the ability to play a song would be expressed in a skill file, right? You might have Eleanor Rigby dot markdown. and then that mom investigated The recent thing I was reading about in one of the Anthropic blogs is...
00:24:09
Speaker
instead of using MCPs and because it bloats a lot of context window of the agency. Yeah, it does. so is Use skills with CLIs, with scripts.
00:24:22
Speaker
Would that be a better approach? That is a good approach as well. I do like running scripts or CLIs or REST APIs. I mean, if there's a, um you have two options, right? If you find a nice Swagger document or open a an open API document, you can,
00:24:38
Speaker
write a skill against it directly and use traditional get CRUD activities. Or you could say convert this to a fast MCP server and abstract away those APIs.
00:24:50
Speaker
I think it's a little bit more intelligent with MCP because you can group various collective APIs together. Like it's sort of, there's an aggregate effect sometimes. I'm not saying convert every single get, post, put, delete directly to an MCP tool.
00:25:06
Speaker
I mean, that's an approach. It's a little clumsy. You'd rather have the ai you know intelligently make your fast MCP server took took to consume those APIs a little bit more intelligently.
00:25:19
Speaker
So there is a there I think there's an advantage to MCP still. But yes, a very so what i what I like about your approach, what you described, is the determinism. yeah So what I'm describing is more reasoning based. Let the LLM reason and pick the tools to run and achieve the task.
00:25:36
Speaker
What you're describing is is a little bit more deterministic and I think maybe more has maybe more appeal in the networking world. Or if I have an Ansible playbook or a Python script. Yeah. yeah This resonated me because I'm more into operations, right? Operation side of things. I think networking included in the same domain.
00:25:57
Speaker
but requires a lot of deterministic approaches. you So what i heard um recently was if you give a hammer to somebody, everything else looks like a nail, right? Yes. yes Instead of solving everything with AI, we could have we we have been doing a lot of automations prior to this AI, right? So why not...
00:26:22
Speaker
build smart or intelligent triggers. Randy was also talking about it. Why not build intelligent triggers and have AI reason and pick those triggers?
00:26:33
Speaker
And then that runs a deterministic code or so some kind of mechanism to do a task. Yeah, I think i um I love the mix, the mix of the two, right? So we can build React agents that... um um You know, we can build react agents that can reason and take action and that action is tool calling that tool calling doesn't have to be MCP.
00:27:00
Speaker
It can be any of the I don't want to call them legacy, but it could be rest API's. It could be. It could even be netconf, resconf. It could be Ansible playbooks, Terraform, Ansible, r Python scripts.
00:27:11
Speaker
I love it. I love the mix of the two, right? So it depends on, like, do I want to... So what I like about the non-determinism is the LLM with those deterministic scripts are the edge cases.
00:27:23
Speaker
So the problem with the deterministic script is that you have to come to every determination, right? Like, if this fails, if this fails, if this fails, if this fails, if this fails, right? Like, you really have to consider that whole item potent pipeline in your script.
00:27:37
Speaker
What I like about LLM's ability is that if it gets a 404, you don't have to write the edge case for the error. It will go, well, what, maybe it's the URL I'm using is wrong, or maybe I need to back off and try again, or maybe it's a, ah you know, a rate limiting thing. It will reason and figure that out. So the mix of having your, you know, deterministic A to B to C to D script with that LLM to handle those edge loop cases that you don't have to write anymore, right?
00:28:08
Speaker
I think that's what's accelerating people is the ability to write deterministic scripts without being held up with every single use case, every single error, or every single possibility. Let the LLM handle that, those cases when they occur, right?
00:28:22
Speaker
um Yeah, it's a bro it's a brilliant approach. Yeah, one more thing that agents can do, John, is write scripts themselves. i eat So if something is being repeated, you know every time you have to turn on a switch on a router, then That's a very you know redundant task that every time the agent is doing via multiple MCP calls.
00:28:47
Speaker
So instead of that, it could come up with a script that does four tasks and then trigger that script occasionally whenever required. So I think a hybrid approach would be like this, a dynamic script, a dynamic deterministic script that gets generated automatically whenever something is redundant.
00:29:09
Speaker
Yeah, I think the ideas are infinite, right? Like I think that's what i love about technology right now is that we can... I often say ai is like a programming language. It's upon the programmers how to utilize it.
00:29:25
Speaker
um Nobody has given us a rule book on... People who built AI also don't know how to use it properly, right? It's an engineering question, yeah. Yeah, that's that's sort of been my sort of approach. And my what I love is that I'm rarely let down when I think to myself, here, let's just take this phone thing, that for example, right? Like, okay, in theory, there's an MCP for Twilio and there's even skills for Twilio.
00:29:53
Speaker
Could I add those to Netclaw? Could I set up a Twilio account with the you know the the introductory 10 bucks they give you to make some calls? And could I let... bi-directionally So like, could I actually say thatt clock and phone me if it detects something worthy of a phone call? And that's sort of the spec I built.
00:30:13
Speaker
And can I phone a number and talk to my net claw to get and have it run things and get information back through my phone? That's a thought exercise. That's it. That sounds like science fiction 10 years ago, five years ago, three years ago.
00:30:29
Speaker
I built an SDD, a spec-driven development, using the spec kit from GitHub. All the specs are in the NetClog code, so you can actually read all the specifications that went into building the vi the Twilio integration.
00:30:44
Speaker
And within, let's call it, be fair, two hours, I was receiving and making calls from an agent that is connected to my network. I was actually able to say, how many CML labs do I have? And it said, here's the three CML labs you have.
00:30:59
Speaker
Can you turn on CML Lab 2? And it turned on CML Lab 2. Now that CML Lab 2 is on, can you help me understand the topology? It has router 1, router 2, switch 1, switch 2. It knew that three of the PCs were turned off because of licensing.
AI's Expanding Influence
00:31:14
Speaker
I'm talking to CML through my agent, of course, but I'm using, you know, like I'm actually talking to it and it's answering me in natural language. So imagine being able to anywhere at any time phone a number and say, hey, what's this you know is everything all right with the network?
00:31:30
Speaker
you know Or like a status update, right? There's some big change going on or you're working through some issue. Call a number and say, is the issue resolved? you know Give me a status update of what's going on. Or it call you, right? Three in the morning, some you know if some malicious traffic is detected on your network or some critical link fails.
00:31:49
Speaker
you'd like you know You'd like to know as soon as possible, right? It used to be the NOC that did this. I worked with Knox my whole life. I had a pager. I had two pagers at one point because there was two different organizations that needed to be able to contact me.
00:32:03
Speaker
So that's all this sort of is. But instead of just picking up a pager that says, you know, major problem, get into the office right away or whatever they could fit on the pager message, right?
00:32:14
Speaker
ah I can take a phone call from the network right away that says, hey, John, the link to Toronto just went down. you know, do you want us to open up a ticket with the ISP? Do you want us to get further investigation? And I could just talk to it.
00:32:25
Speaker
It's not a person. It's always there. it can go I don't know. Am I crazy about this? Like, does this resonate with you being able to phone your network? Like, am I insane? Does this sound like, really, though? Tell me.
00:32:38
Speaker
have the have Be honest with me, Priyank. Say to me right now, knowing like, you you've lost it, John. This is too much. see I love one thing about you, John. I was listening to one of your podcasts earlier and you were so excited to tell people regarding, you know, go to Gemini, build build an application. it it can do this. You can build a website. I was so fascinated. Like...
00:33:02
Speaker
People who are very new to the space are not aware. For me, it might be like, okay, yes, I can do that. I already know it. So it it may not seem very exciting for me because I already have felt it once. right But for somebody who's very new, this enthusiasm is something that will drive them crazy and they will come and start picking it out.
00:33:24
Speaker
So thanks for doing that, by the episode Well, like I said, I've been having a lot of fun with this NetCloud project. In the VibeOps forum, we kick around ideas.
00:33:37
Speaker
We tied it into Twitter, and it's actually able I'm actually able to kick off work. So I said to tweet in a tweet, can I get a mind map of my CML environment? And I sent that tweet, and the NetCloud sent a reply to that tweet with an image of the CML in a mark map, mind map for me.
00:33:54
Speaker
Within a couple seconds through Twitter, I'm able to talk to my network through Twitter. Now, why would you ever want to do that? Would you do that in production? Probably not, but it's a fun art of the possible home lab kind of thing. Imagine your home lab being able to send tweets to your home lab or whatever.
00:34:10
Speaker
Now, imagine if I could say it through a tweet. I'll be home in 20 minutes, turn on the lights, you know, open the garage door and put on the Sopranos on the TV with ah with a smart home.
00:34:22
Speaker
And I don't know, who knows, right? It opened up a multi-channel way of interface with with your in systems, right? um Earlier, it was just a you remote control that you would use to turn on and off TV. Now it's much more than that.
00:34:42
Speaker
You can literally talk to your television now. Right? Right. Right. and This is what i started with, John. in In Rubik's Cube as well, the ah first feature that I built was talk to your infra.
00:34:58
Speaker
i Let me see. Cool. cool yes
00:35:04
Speaker
Do you see this? Yeah, talk to your intro. Hey, I have to get you one of my stickers. i go You need my face on your laptop. Do you know what I think the next switch is?
00:35:15
Speaker
You are in the modem generation between pre and post AI. know right the malaios actually and look you here What does look like to you, John?
00:35:27
Speaker
Well, I'm just saying that there's going to be a generation that are born into a world that's post-AI, and AI is in everything and everywhere, right? So history repeats itself. like I went from analog to digital to AI late in my life, whereas you started in digital to AI. There's going to be a generation that's born in AI, and what's after AI? Maybe quantum?
00:35:49
Speaker
What's the next big thing that we break or or figure out, right? So post-AI, I see it absolutely everywhere, Priyanka. Like I see absolutely everywhere across humanity, much like the World Wide Web. So every company that has a website today is going to have an agentic system. They might have hundreds of thousands of agents instead of a single website.
00:36:11
Speaker
Hospitals, agriculture, manufacturing, all the places we talked about, all the different, finance, trading, everything is going to rely on agents. Humans are going to start to rely on agents, much like we rely on honeybees,
00:36:25
Speaker
much we like we rely on our automobiles, much like we rely on our cell phones and our televisions. We're going to come to have a reliance where tokens are a new currency. Tokenomics is a real thing.
00:36:39
Speaker
Open source models, private models, Chinese models, Japanese models. That's the post-AI world is is is keeping up with the latest model, ah the physical man of state manifestation of AI is coming.
00:36:53
Speaker
I was around at Google Next. I was near some of those Boston robotics dogs. Have you seen those? Do you know I'm talking about? I was two feet away from one, and it gave me the creepiest feeling. like I felt like I needed to run away.
00:37:07
Speaker
i had a real fight-or-flight sensation. like like It was really weird. like I knew that that thing could kill me, and I would have no defense against it. because if it's you know it's a strong robotic dog. Like how would you stop that, right?
00:37:20
Speaker
So anyway, that was kind of neat. But yeah, the physical manifestation, i what I hope, I hope it leads to things like augmenting ah the life experience for people with disabilities, for the blind, for the physically impaired, for the deaf, for the hearing impaired, for the you know aging population.
00:37:42
Speaker
For the young children, like I you know i was reading Bill Gates a while ago, he said that that AI is going to help humans become literate by the time they're the age of three. imagine the Imagine the advancements in humanity if that if that holds true.
00:37:57
Speaker
If our literacy moves from five or six or however, whatever the average age is now to three, because we can give our children access to this AI that will teach them how to read and be literate,
00:38:11
Speaker
um You know, imagine the the compounded interest on that over time, that the human race has started to be able to read younger and younger and that people have access to infinite knowledge at their fingertips.
00:38:24
Speaker
Why is the sky blue? do you know many parents are tortured by that question? By their children? Well, now you can actually help in their terms, help them understand it with an AI. You know, i think that's part of it. I think we need to maybe set aside our ego and our hubris that we're still the dominant intelligent life form on earth in terms of brain power.
00:38:47
Speaker
Like I wouldn't, I would rather have an AI explain that to my niece or nephew than me try to explain it. I don't know. i mean, I sort of know why the sky's blue, but in terms of institutional, you know, generational knowledge, wouldn't it be great to be able to teach our children truth, science-based facts about things, right? Imagine that impact around the world.
00:39:10
Speaker
you know I don't mean to get religious or political here, but why do you think the Pope has come out against AI? Because AI is going to be the final the final thread that breaks and and and and gets rid of the foothold of religion around Earth.
00:39:27
Speaker
Because the AI is going to give us science, atheistic-based answers. right They're not trained on religious data. right So I think that that is impressive.
00:39:40
Speaker
They are partially John, by the way. Are they? Well, that's unfortunate. Maybe I need to train a new model that's purely atheistic. There's one question I would like to ah come to, John. is do you think So you're very optimistic about ah the potentials and the hopefulness that I see with you is with AI.
00:40:03
Speaker
Do you think we have reached reached the zenith of the development or do you think this LLM thing is going to be the end of AI evolution or something else?
00:40:17
Speaker
Well, i I mean, I try to look at the patterns in history and this is our this is going to be our worst attempt at it because it's our first attempt at it, right? So I think it's only going to get better and I think it's only gonna get cheaper and faster and more available and and more quality answers over time.
AI's Role in Healthcare and Society
00:40:36
Speaker
I don't think we've reached the Zenith um because then the next thing will come out, right? that there There will be improvements on this. The other thing is like the US government has asked OpenAI to not release ChatGPT 5.6.
00:40:50
Speaker
Why would they do that? what You know, like, is it so powerful and so capable that the governments of the world are asking the AI companies now to not release models.
00:41:03
Speaker
mean, think of that. Is it because it's because it's that powerful that the governments of the world don't want us to have it yet? um So it's – I still – here's one thing I do, Priyank, and I mean this. I do this every week to start my week.
00:41:20
Speaker
I search for Google. I do a Google news search for AI cancer, ah you know, news or AI cancer research or AI cancer breakthroughs.
00:41:31
Speaker
I do that once a week and I've done that for a while now. And some of the things I read are so hopeful and so optimistic. um You know, I, ah to make it a little more personal, I had a colonoscopy done not too long ago. And apparently while I was under the drugs,
00:41:47
Speaker
All I could ramble about was artificial intelligence. but So the nurse told me after, after I had kind of, you know, and I was fine by the way, it was just a, it was a routine thing. um She said, actually I did some research and, and, and the results are being scanned by and private age ai and false positives have gone down 70% since they implemented this AI to look at colonoscopy imagery.
00:42:16
Speaker
right? Yeah. Like 70%. So that's, he said that's 70 people out of a hundred that were not telling them that they're positive for cancer. Right. So whether or not it cures it, I don't, you know, obviously that would be ideal, but what if it can help in those areas, right. Or, or make drastic improvements in, uh, prevention or, you know,
00:42:39
Speaker
ah you know maybe not curing it, but helping remediate it I don't know. I'm optimistic. i saw it such So to do that, if AI gets you down and all the doom and gloom ah bothers you about AI, try to search for positive news stories about AI. Try to find the helpers. Try to find the people that are trying to use this technology to save lives, to make life better on Earth, right?
00:43:03
Speaker
i mean, I'm doing what I can with network automation and engineering. I'm i'm trying to to use the tools to make that thing that we do a little easier, right? So I can only imagine, I hope there's people like me in other fields around the world trying to implement AI to make their field even better and more augmented and AI enhanced and all of that, right?
00:43:26
Speaker
you You touched upon a very deep points, John, and I totally am aligned. I fully embrace the new tech, new developments in science, be it AI, be it quantum, anything that comes. up Because if you do not embrace and restrain yourself from adopting new things, then You are in bad luck, that's all.
00:43:57
Speaker
Things will keep developing. There is no stopping that. It's just that how well are you going to adapt to it. the more you push i away yourself from it, the more time you will take to catch up ah to the next stages. i e I liked your perception of yeah just just a small ritual, right? Every Monday, just go and search for cancer cures using AI. And maybe someday we'll have that.
00:44:29
Speaker
Well, we can only hope so. Or or natural disaster detection or better detection. mileage in our gas, you know, anything, anything to improve the human experience. We use this technology to to actually advance us forward and and make, you know, bring up the quality of life. And and I like to think that um I know it might not seem that way for everyone on earth, and I know I'm very privileged where I am, but if you look at the numbers globally, right, poverty is down, ah for infant mortality is down, life expectancy is up, quality of life is up.
00:45:03
Speaker
I know it's hard to accept that, but statistically speaking and more broadly speaking, right, women around the world have more rights than ever. You know what I mean? Like there's a lot of positive things that we just take for granted or or don't really think about.
00:45:18
Speaker
I also like to walk my dogs in the forest and not think about AI at all not think about technology at all. I have a forest perspective. cross-country skiing paths that that are allowed to walk your dogs on in the summer.
00:45:32
Speaker
So I will contemplate a lot of this work just walking through the forest with the dogs. But, um you know, i ah that's my, that's that's, what else are we going to say? That that that it's going to be worse for humanity? That it's going to make life more difficult?
00:45:47
Speaker
I know that there's a lot of... Yeah, this, how do I phrase it? Being skeptic, being a bit ah concerned regarding what could be the implications is good, but straight away denying things is not at all.
00:46:07
Speaker
um There are a lot of good things that are happening. We have to be safe. We have to be you know conscious about what new models are coming. Maybe they are very advanced. Maybe it's a marketing gimmick. Who knows? right So things are like that. ah It's the judgment that we have to always take into consideration about anything that we develop.
00:46:30
Speaker
that's That's where I think the judgment of humans always will come in. This is, Randy was saying, for designers, for developers, The taste is something you cannot replace with AI.
00:46:43
Speaker
Same thing for operators is you cannot replace human judgment. I think this applies for everything. This taste and judgment is some of the very innate human qualities that you cannot replace with ai And there's something which will lead us forward.
00:47:01
Speaker
Well, you know, I think some people have to experience it firsthand. I really do. And then that's and that's the biggest hurdle. That to me is the biggest hurdle is taking that very first step. And I see it almost every day, people you know celebrating their very first step.
00:47:17
Speaker
I just made a pull request to my first Git repo. I just did my first Git clone. I just integrated my first MCP server. Now, I really want to stress that for anyone watching this. If you haven't an integrated an MCP server, you I think you're falling behind. I think you fell behind. i think you need to do it. I think you need to try it.
00:47:37
Speaker
It will take you 20 minutes to integrate, say, the netbox read-only MCP server. I've covered this publicly many times. ah You really need to do it. And and you're going to see the light bulb is going to go off and you're going to understand. And then I think your next step is going to be to find more MCPs that apply to your life and the tools that you use, right? Because you're going to be pleasantly surprised to find that there's likely an MCP server for the system that you integrate with and use every day.
00:48:05
Speaker
Your next step is to get approval or whatever you need to do, access to a local model or and a company approved model and connect that MCP to Claude Code, Claude Desktop, Copilot, Antigravity, whatever it is you're allowed to use, Cursor.
00:48:19
Speaker
and And you'll never look back. You'll look back and say, why did it take me this long to do this? I see what they're talking about now. Holy cow, is this ever powerful. I mean it. i you know It's undeniable.
00:48:32
Speaker
I've yet to meet a person to try it and go, well, I'm going to do it the old way. Like that's, this is no good. You, you, you would be surprised. So how many people have just have a mental block and, and either they just, you know, like I said, there's some hubris, there's some ego, there's some personal emotion, whatever it is that's holding them back.
00:48:50
Speaker
It's 20 minutes of your life to install an MCP server and try it. And I, so I promise you, I, you reach out to me and tell me otherwise. Say I tried it, John. And, uh, It took an hour or two hours instead of 20 minutes. That's not going to happen. And once I got it working, it was awful or terrible. That's not going to happen either. Like it never, ever happens. pream You know yourself, right?
00:49:13
Speaker
you've You've seen people converted to this and and and it's a like it's like a flipping a bit from a zero to a one. You're, you're, you know, augment yourself with AI. If you've tried chat GPT and you got some bum answers or asked that how many hours are in strawberry or whatever it is, that's not using AI.
00:49:31
Speaker
That's a toy that's consumer garbage. That's just, you know, you're skimming the surface. If you've had a 20 minutes with a chat interface and it didn't go the way you think it did, uh, rethink your position on AI really though, really take a step back and, uh, and can reconsider using these tools.
00:49:50
Speaker
in In a year, it's I don't know. It sounds silly to even say this in 2026. I still can't believe people aren't doing it this way all the time. you Maybe in three years or five years or two years. People take adopt. We're going to look back at this and laugh that we even have to encourage people to get started, right?
00:50:06
Speaker
People take time to adopt, John. And the pace at which AI is moving is too fast for... you know regular people to adopt.
00:50:17
Speaker
And that's the problem. That's the challenge I'm seeing.
Overcoming Fear and Embracing Innovation
00:50:20
Speaker
It's not that they don't want to. ice but yeah So it's very interesting. um I was at SRE day just last last Sunday, last Saturday ah in Bangalore.
00:50:33
Speaker
And I gave a talk on fear paradox for SREs. And I call it fear paradox it because there are two fears. One is if I don't adopt AI, I'm going to lag behind. That's a fear.
00:50:51
Speaker
So I must adopt it. right For SREs and operators, another fear is if I adopt AI, then there are lot of problems. You cannot run it in production. There will be issues. Something may break. This is another kind of fear.
00:51:07
Speaker
So whether to choose it or not choose it, both are the fears that SREs have today. And this is where I broke it down. whites know The fear itself is a spec, a design spec, a PRD for what you need to build.
00:51:22
Speaker
right And this is where I think the most important point was that the Fear, the the reluctance of adopting new things is just because of speed, because of that empty canvas feeling. Like you you have an empty canvas and you do not know what first stroke should be. What to paint. Yeah, I, yes, I completely agree with that.
00:51:46
Speaker
Well, that's, you know, so you asked me a little earlier about my motivation, right? And I really would like my my videos in particular, my body of work to help give people something like the Bob Ross experience.
00:52:02
Speaker
right yeah So i was water come yeah they can watch the episode and watch me paint something and the mistakes, you know, Bob Ross with the happy little mistakes.
00:52:13
Speaker
He doesn't, you know, edit that out. He, he leaves it in. He shows how to paint around it or how to correct the mistakes. You know, I'm not trying to say I'm Bob Ross or anything like that, but ah to your point of some people have a blank canvas and they just, so maybe they could have their blank canvas and put me on in the background and try to replicate what I'm painting, right?
00:52:32
Speaker
Meaning, or or their own inspiration. They can watch me build it and then go, oh, I see, I get it I'm going to apply what I just watched to my own project, my own idea.
00:52:42
Speaker
And that's what excites me the most. It's, it's, it's the people that like yourself, what did you tie the phone into? What was the system you had mentioned? Rubik's cube, right? I didn't do that. That was your inspiration and where you took the idea, right? That's what's exciting to me. And that's how these, these neural networks of people and, and connections and,
00:53:05
Speaker
You know, I'm never going to forget this interview. I'm never going to forget meeting you and and knowing that I've inspired you to build something and you'll likely show other people. in And it's not about credit or ego. It's it's about community and and feeling good about myself. and And I get really good endorphins. I really get really good neural feedback.
00:53:24
Speaker
When I hear people say you helped me or or this was great or whatever. Right. So that, you know, I don't do it because of the money. it's not like I'm, I'm not one of the big, you know, million subscribers or 70,000, you know, I've got, I've got about 8,500 subscribers on YouTube.
00:53:39
Speaker
um you know, I get a lot of people who do watch the videos, hundreds of thousands of views, which is incredible. And I thank everyone who supported me. Um, but it's certainly not for likes or clicks or money that, that I do this. It really is a,
00:53:54
Speaker
It's more about legacy than anything. I'd like, you know, when I retire or when i look when people look back at my career or the industry that I've i've made some
Spec-Driven Development and Final Thoughts
00:54:03
Speaker
impact on it. I've had a um a small mark on it that people will say, oh yeah, he was the first guy to make a phone call from the network, right? Little things like that. i i thought if That's what excites me.
00:54:14
Speaker
But in terms of the way I've been building things, this is what excites me the most that you can be me. Honestly, like I can't stress enough how empowering the new spec kit from GitHub is and Microsoft and how powerful spec-driven development is.
00:54:30
Speaker
I need you to go and and look at this, everyone, because you you it's like Git in a way. it will instantiate and and you still have a working directory, John's john's video game.
00:54:41
Speaker
Okay. You instantiate it for the spec. kit and that will bring down some templates and some other scripts and some things like that then you just follow the six step process first step is a constitution second step is your specifications there's an optional step of clarification to make sure that your spec and your constitution align or that the llm doesn't have any questions so far then you do spec kit plan Then you do tasks.
00:55:09
Speaker
Finally, you do implement. Now the first bunch of steps all create markdown files. Your specifications. These are user stories. Priyank, you'd love it. You're going to love this because they're user stories, they're pass-fail acceptance criterias, they're functional requirements, there's test cases.
00:55:29
Speaker
Everything is documented in markdown. And ultimately you get to your tasks and it's X number of tasks. When you say implement, It builds your code, your Python, your markdown spec, your Ansible playbook, your blog.
00:55:45
Speaker
Anything can be done this way. And what's neat is that when you want to look at my work and say, well, how did he build this? How did he come up with it? You can look at the specs. You just read the specs. It's all in the Git, right? So then if I want to say add multiplayer to John's video game,
00:56:00
Speaker
I just started the spec step. I don't need to make a new constitution. And it will track a new folder with all the artifacts for your second spec. So over the lifetime of a project, if all you do spec-driven development, you're going to have a spec folder with every branch that reflects a spec that you've added to the code base.
00:56:22
Speaker
It is remarkable. I cannot say enough about this approach. um It's how I build Netclaw. So the Twilio thing. This is one of the like things that people, you know, in the industry, we have this term of knowledge transfer, right?
00:56:39
Speaker
I can do a lot of things, but i if I'm capable of training another person to do the same thing or maybe more advanced thing in the process of how I do is very interesting. And and the way you have explained, I'm just definitely going to check it out.
00:56:56
Speaker
I think this KT thing, the knowledge transfer things, is automatically the done now. thank to you However you did the operations is automatically documented and nope the next person can know about it, can rerun the same thing and get the same results.
00:57:14
Speaker
Right. Well, you're you really are going love it. and I think your listeners are going to love this approach if you haven't heard of it. um yeah you you've probably done tdd right test driven development edgerman yes where you write a failed test and then you make the minimal amount of code to make it pass and then you iterate and and refactor right so if you've done tdd you're going to love sdd even more it's going to feel very close to your heart we we covered a lot john and we We talked about you know the excitement that you have, the the hopefulness, the know the enthusiasm that you are seeing in the industry with the changes that are coming in, with the
00:57:58
Speaker
ah the way humanity is driving forward. I think this is very inspiring. Your life journey itself is very inspiring. And I think I can categorize you as this Bob Ross of the tech world.
00:58:14
Speaker
um And this is this is going to be wonderful, Ron. This episode is very interesting. People are going to love it. Any message for my audience before we close it?
00:58:28
Speaker
Yeah, ah so I mean what I say in that you can do this. If you listening are watching, install the spec kit. And here's a fun thing. Make a video game about something you're learning.
00:58:41
Speaker
So if you're learning about OSPF, if you're learning about BGP, if you're learning about COS, if you're learning about gardening, if you're learning about whatever, any topic at all, in your constitution, you're going to explain what to do and what not to do.
00:58:54
Speaker
And you're going to explain the guardrails, like the American constitution. Think of it that way. You're making a constitution for this project. Your spec, you're going to say, I'd like to build a gardening video game that's educational to help me learn about weeds and plants and seeds and whatever.
00:59:09
Speaker
And you just go through the process. After you do slash implement, you will get a working video game in your browser. I've done this. I've made Solitaire. I've made network automation video games. Someone at Cisco Live made a hot aisle cold aisle video game. Someone else made a BGP shooter.
00:59:26
Speaker
So you don't have to be a video game developer. You don't even have to be an expert in the topic. Use Spectre-driven development and build yourself a fun little video game. And also, feel free to reach out. Make sure you join Priyank and I in the Vibe Ops forum.
00:59:41
Speaker
We're always there to answer your questions and to explore ideas. It's an open, safe space. And Pian, you've said some really kind things and really encouraging things to me today. And it really I really feel it and and it means a lot to me. And and I look forward to the next time I'm on the show.
00:59:58
Speaker
Absolutely, John. And thanks thanks a lot for taking out the time. And thanks a lot for addressing my audiences. um We'll catch up soon, John, whenever you are ready.
01:00:12
Speaker
Thanks a lot. Bye-bye.