How John Peebles Used Claude Code to Turn Manual Work Into AI Automation

Introduction & AI Sherpa Series Launch (00:01)
Julian introduces the first AI Sherpa episode and welcomes John Peebles, CEO of Administrate.

What Administrate Does (02:20)
Enterprise platform managing complex, global classroom training logistics.

AI Skeptic to Practitioner (06:19)
John shares his journey from AI skepticism to experimentation after Claude Code’s release.

The “Pyramid of Suck” Problem (10:18)
Manual, error-prone enterprise configuration during customer onboarding was constraining growth.

Legacy Code Meets AI (16:24)
Using Claude to interpret and programmatically manipulate legacy configuration systems.

From Spreadsheets to Automated Setup (20:15)
AI-generated decision trees and config scripts replaced dozens of hours of manual setup.

Sales Handoff Automation (24:57)
Pulling Gong transcripts into structured, referenced implementation documents.

Customer Impact (27:31)
Faster onboarding, higher quality, improved perception, and easy reconfiguration.

Internal Adoption Challenge (29:59)
The hardest part wasn’t the tech—it was getting people to believe it worked.

Advice for Leaders (30:35)
Experiment personally, start with coding use cases, focus on low-risk internal automation.

Why Code Is the Sweet Spot (31:05)
Code can be tested and validated—reducing hallucination risk compared to general AI tasks.

Closing Reflection (37:44)
AI experimentation unlocks practical leverage when leaders get hands-on.

Julian Castelli (00:01.584)
Good morning. This is Julian Castelli and welcome to a special episode of the Growth Elevated Leadership Podcast. This is our first AI Sherpa podcast where we talk with tech leaders and we’re going to talk about how they’re using AI to succeed at their business. This is part of the broader Growth Elevated Leadership Podcast where past guests have included CEOs and CXOs of great companies like Workfront, CHG Healthcare, Pathology Watch, InMoment.

Canopy, the San Francisco 49ers, and many more. This episode is brought to you by Growth Elevated. We are a community of tech founders, CEOs, and CXOs who are committed to working together to share best practices and learnings in an effort to help all of us become better leaders. We do this through educational programs like this podcast, as well as our blog, and of course, our annual Ski and Tech Summit, where we bring tech leaders together in beautiful Park City, Utah, from all over the world.

So if you’re into skiing and tech leadership, check us out at growthelevated.com. Today, I’m happy to welcome John Peebles. John is the CEO of Administrate. Administrate is a leading training management system that helps large enterprise customers plan, schedule, communicate, and adjust their training programs with their employees and their customers. John is a technical founder who previously served as a VP of Technology and CIO.

for several healthcare technology companies. John is currently a board member of Current Health and TuringFest. John is coming to us today from Edinburgh, Scotland. Welcome, John.

John Peebles (01:37.432)
Well, thanks so much for having me. It seems like whenever you’re involved in Scotland somewhat, the weather is beautiful outside. So when you visit, it’s beautiful. Today is a beautiful day. yeah, that’s that’s actually the case. Yeah, no, it is. It’s always beautiful. Thanks for having me.

Julian Castelli (01:47.123)
I thought it was just always beautiful. It’s not that that not the case Yeah, I’ve enjoyed I’ve enjoyed our visits I want to do it again. So John, this is super exciting I know you’ve been working a lot with AI and and you’re running a tech company but but we’re at the intersection where tech companies are Using a new technology and you’re particularly well suited to talk to us about it giving you a technical background

Before we get into it, tell us a little bit about administrate. What does administrate do?

John Peebles (02:20.024)
Yeah, so Administrate is a platform that is designed for usually large enterprises to manage all of their classroom training. Classroom training, believe it or not, is still pretty much the predominant way that training gets delivered in corporate America today, particularly in highly skilled or highly regulated or environments that somebody could die or millions of dollars of equipment could be lost. And so there’s no better way to get a student to a classroom. The problem with that is that it…

Julian Castelli (02:44.143)
And so it’s a way to put a classroom to the problem with that. It doesn’t shoulder it well. people look at certain places at certain times and all that logistics. That’s horrible. We’re really trying to keep the trend the same. I think that’s the spirit of this. It’s not my idea. The United States is free. Our customers are operating across the world. Our thought for basically helps identify where it is, where there might be problems.

John Peebles (02:49.41)
doesn’t scale very well and you got to get people in certain places at certain times and all of that logistics is a struggle, particularly when you’re trying to keep the training the same in China as it is in Europe, as in South America, as in the United States, which many of our customers are operating across the world. So our software basically helps identify areas where there might be problems, alerts our users to those areas and also helps them significantly in the planning and scheduling phases, which

Julian Castelli (03:13.531)
Sounds like a huge logistical challenge if it’s in person, right? I thought I thought it was all moving to the LMS’s. That’s not the case,

John Peebles (03:18.326)
You can imagine having to plan thousands of classes globally for maybe a quarter or two quarters ahead can be a really big lift. So it’s an interesting problem to solve. Yeah, it’s not, yeah.

No, I mean, there’s been millions, hundreds of millions of dollars spent in trying to convince everybody that all learning will be done online and self-paced and all that. But I think, you know, classroom training is here to stay. And the pandemic was a big test for that. think training levels in classroom stayed unchanged, although they went to more virtual means. But yeah, we we partner with a lot of the traditional learning management systems out there. And it’s just a great way to give

learning leaders, multiple ways to tackle a training problem, whether it’s self-paced in classroom, a combination of both, and we can drive costs down on the delivery there on the classroom side.

Julian Castelli (04:15.522)
Terrific. so your customers are large enterprise companies and they have to bring these people together. So it’s like a logistical planning tool set that allows them to schedule that almost like a school administrative system, right?

John Peebles (04:30.54)
Yeah, totally. mean, look, we’ve all spent lots of time in school, but we very rarely think about the logistics involved in getting that classroom and the materials and the teacher and the students are ever in the right place at the right time. And I would say probably the closest analog to the scale of this problem is we’ve all probably flown on an airplane and maybe had a scheduling mishap or something along those lines. It’s a lot like that for our customers because they have

Julian Castelli (04:42.945)
Right.

Julian Castelli (04:50.842)
all probably the form of an airplane, maybe not a sketch, maybe a job or something along those lines. A lot like that for our customers because they have these sort of boxes that people can sit in the environment, their equipment, and then things happen, weather happens. If the equipment breaks down, so your flight destiny is scheduled, and if the exact problem of my flight to university is a bit more difficult, there’s a problem.

John Peebles (04:59.016)
these planes, which are classes that need to run and people need to be at the right place. You have the right equipment and so forth. And then things happen. Weather happens and equipment breaks down. And so your flight has to be rescheduled. And it’s the same exact problem of, no, my flight to New York City has been rescheduled or there’s a problem and we’ve put you on this other plane. And if you don’t like that, give us a call. It’s a similar experience when you think about classroom training.

Julian Castelli (05:20.729)
We put you on the separate line. you don’t like that, this is a other experience. Think about classroom training. If one of our customers has a different track, for example, the Monopolis, if a lot of allies have a program that we have to get in, the way the student provides the machine, we need to flat-touching. So the free involves everything in the machine.

John Peebles (05:27.71)
And some of our customers are doing all kinds of things that you wouldn’t expect. Roche is a great example on Indianapolis. They have these blood analyzer machines and you have to learn how to operate them. The FDA regulates these. You can buy the machine, you legally cannot touch it until you’ve gone through training. And so the training involves everything from getting the machine in the right place to making sure that the fake blood is at the right temperature in advance. And you can run the fake blood through the machine. You have to clean it afterwards and…

These types of things we just don’t think about when we’re thinking about the education requirements in an industry like healthcare, for example.

Julian Castelli (06:04.272)
Terrific. And so what’s been your company’s journey with AI? We’re going to talk about one of the specific projects that you’ve had success with today, but

What’s been the journey in terms of how the company’s been adopting AI? me where you are on that journey.

John Peebles (06:19.905)
Yeah, so I mean, one thing that I think is really important to understand is I am an AI skeptic. I, when, you my, I’ve got a computer science degree, special, specialized in AI a long time ago, 20 plus years ago, but a lot of the techniques and concepts remain. And when LLMs kind of busted out on the scene, you know, like everybody, tried it, you know,

Julian Castelli (06:25.14)
different.

John Peebles (06:47.405)
tried to ask how many Rs were in strawberry and all the little gotchas and things and I was pretty underwhelmed. It was cool to be able to generate some weird AI slop art but apart from that, I didn’t personally find it very meaningful in my day to day life but I found myself at this point about a year or so ago, actually the last two and a half years ago where I was convinced I was right and this was mostly hype and not very useful.

Julian Castelli (06:55.441)
So you dug yourself in as a skeptic.

John Peebles (07:17.301)
And yet everybody else is convinced that I was wrong and they were right. you know, when you’ve got, yeah. And, you know, it’s like, it’s like, I would go out for beers with like, there’s a, there’s a real good friend of mine lives across the street, building an AI company, sold a previous AI company to Facebook, now Metta. And I’d just be like, am I insane? You know, like, like I’m just not finding this to be that useful or whatever.

Julian Castelli (07:33.505)
I just feel like I’m just not finding the disc to be that musical or whatever.

John Peebles (07:44.55)
And you know, there’d be bits and pieces and he’s a more nuanced, measured, much more intelligent guy. And he would kind of point out, yeah, but these are fringe benefits. It’s not this cataclysmic thing. Until about, but I think one of the key things like in business, know, startups, you’re looking for product market fit. You you cannot just write off what you did two years ago because we tried that already. You got to keep going back and revalidating all the time.

Julian Castelli (08:12.556)
Because it’s changing so fast, right?

John Peebles (08:14.421)
It’s changing so fast and it was about almost a year ago in May of last year when Claude Code got launched. And I started playing with this like I would and was very skeptical and prepared to be underwhelmed. Trying, yeah, absolutely. you know, trying to torture this thing and get a gotcha moment or two and whatever. And the interesting thing was that I was like, wow, this is actually doing stuff.

Julian Castelli (08:23.053)
Right.

Julian Castelli (08:29.537)
Trying to break it, trying to prove your thesis.

John Peebles (08:44.235)
that is useful, that is not terrible. And that kicked off basically a year plus now of experimentation. And it wasn’t just me in a vacuum either, but it was, again, colleagues, folks that I’ve met over the years, all trying to play with this thing. And that’s kind of how it got started for real. was about a year ago.

Julian Castelli (08:50.25)
it costs basically.

Julian Castelli (09:02.328)
That’s kind of how that started for real. Terrific. Well, listen, we’ve got this framework that we’ve been talking about at Growth Elevated of three different levels of AI implementation utilization. The first is to enable your employees. The second is to automate some of your internal functions that are some of the most manual, difficult things. The pyramid of suck.

And then the third is to actually put AI into your product. I think you’re gonna share an example of stage two with us, right? You automated an internal process, is that the case that we wanna go walk through today?

John Peebles (09:37.526)
Yeah.

I think so. think it’s an example that I find to be very actionable, very practical. it’s something if you’re listening, you can try out with fairly low barrier to entry and fairly low blast radius. And you can probably also try it out yourself if the rest of your company is not believing any of this AI hype like I was. Yeah.

Julian Castelli (09:53.015)
Perfect.

Julian Castelli (10:01.89)
Well, sometimes you want to prove it through an example, right? And that’s why I wanted to share the story. So tell us what was the process and what was the process before? What put it at the bottom of the pyramid of suck that made you say, man, this is worth trying to shine a laser beam at?

John Peebles (10:18.903)
Yeah, so we have just come through an interesting time at Administrate. We kind of effectively rebooted our entire market focus, market segment, go-to-market strategy, entirely new commercial team came in as we now today only sell to large complex enterprises. We used to have a business that was SMB focused and that’s kind of gone by the wayside. So I feel like I’ve been, Yeah, I feel like.

Julian Castelli (10:42.807)
And that’s a huge transition. It’s a lot more different than you’d wish,

John Peebles (10:47.767)
feel like I’ve been the CEO of two companies, maybe even more. it keeps it interesting, but it is extremely difficult, took longer than anybody thought, was more expensive, et cetera. As part of that, your implementation efforts really ramp up in terms of the complexity.

Julian Castelli (10:52.652)
If you just.

Julian Castelli (11:04.253)
That’s right. Enterprise customers expect more.

John Peebles (11:07.467)
The change management just on the people side, the configuration that has to happen, the integrations that they are looking for with other systems within their tech stack, it’s all just a completely different beast. And for us, that beast we were hoping would be about 12 weeks, which is kind of the industry standard, at least in my experience. But you always want it to be shorter. The problem that we had was that our platform

Julian Castelli (11:18.615)
It’s all just.

Julian Castelli (11:27.607)
Yeah.

John Peebles (11:37.045)
requires a number of decisions to be made kind of early on about how it’s going to be deployed. So just simple things like, are you selling training or not? Right? If you’re selling training that opens up a whole can of worms on discussions around what are the business models you’re using and are you going to use our infrastructure to sell it online or do the orders come in from Salesforce or do they come in from both your website and Salesforce? All of these types of discussions.

Julian Castelli (11:57.036)
So all the configuration steps that you might need.

Yeah.

John Peebles (12:06.476)
And this is the type of thing that would take a long time during the implementation process to not only get through from just a dialogue perspective, but once we had the information, it’s like, okay, yeah, we sell in 25 countries and of those five are franchise, you know, sales channels where we don’t really control everything, but we want to know what’s happened. And, you know, so these complexities start to balloon out.

Julian Castelli (12:31.305)
And where would these complexities go to? Like what was the process? Would you have an implementation manager and a spreadsheet basically?

John Peebles (12:38.156)
Yep, yeah, the team was a project manager and then a solution consultant and then a bunch of wild spreadsheets. And that was just for the configuration information. And then you would get to the point where it’s time to load data. And this is a familiar problem for a lot of us in enterprise software, but you’d then get the data and it would be wrong in many different subtle ways and you’d or incomplete or somebody would be

Julian Castelli (12:58.793)
then if it does, would be.

or incomplete.

John Peebles (13:07.062)
promising it any day now, because they were typing it up from whiteboards and whatever. And then you’d have to get that loaded in, test it all with the configuration, and then you would hook up various things and hopefully go live.

Julian Castelli (13:20.371)
You’ve described the product as almost a Lego set. Like you can configure it in so many different ways, which is a strength. At the same time, the complexity that that creates seem like it almost can be overwhelming.

John Peebles (13:31.338)
Yeah, 100%. And this was a problem for us because our go-to-market motion was starting to work in some ways better than we anticipated. We got some more complex customers than we expected. And we were projecting out and saying, look, we’ve got a small team. We want to keep it small. We don’t want to get out over the skis. But we’ve got some large customers that are in the pipe that are maybe out for signature. Yeah.

Julian Castelli (13:58.229)
Large customers, large expectations,

John Peebles (14:01.227)
Yeah, and so then you extrapolate out, okay, it’s gonna take, and so this configuration information, which should be stored in a spreadsheet, would then have to be translated into a control panel. And if you look at our control panel, it’s like a lot of enterprise software, there’s like 50 or 60 different sections, and each of those have tons of knobs and things, and yeah. you know, what, and like,

Julian Castelli (14:22.013)
like flying a plane, a complex plane.

John Peebles (14:28.736)
We weren’t idiots necessarily. We understood that this was a problem and our implementation team would come and say, hey, twisting all these knobs and fiddling with all these dials and translating the spreadsheet into the configuration control surface was time consuming and manual and error prone. And if you got something wrong, sometimes it’d be difficult to go back and start over.

Julian Castelli (14:45.022)
whole surface.

that figure something in. Sounds like a perfect example to try to try to automate or do something with AI because time consuming, complex, manual, important, and repetitive and growing if your sales are growing well. So you saw this big.

John Peebles (14:57.078)
configure something in a weird way, the data model configuration, whatever it would mean, you’d have to almost start over in some cases and redo that bit of setup.

John Peebles (15:15.424)
Yeah.

John Peebles (15:22.016)
Yeah.

Julian Castelli (15:24.392)
growing body of work that was going to constrain your growth?

John Peebles (15:27.87)
And you extrapolate it out and it’s like, we cannot have a two person team be doing one to three projects at a time. know, that there’s just not, it’s just not, yeah. And so the challenge is you, this team would come to our product meetings and say, Hey, we need to be able to script this or run things in programmatically. And our product team would be like, well,

Julian Castelli (15:35.784)
Yeah, it won’t work.

John Peebles (15:55.275)
Yeah, I mean, this control panel was largely built on legacy tech and we’ve been investing lots and lots of effort and time on building a really nice modern experience. But like every company, we have legacy stuff. And the problem is, is that you only set the product up once or twice, or you fiddle with it maybe once a year or every 18 months or 24 months, right? Yeah, the volatility of the configuration changes is low. I mean, it’s very, very low. And so…

Julian Castelli (16:10.34)
Yeah, once you’re done, you don’t want to touch it again.

Julian Castelli (16:23.278)
very, very low.

John Peebles (16:24.628)
you could never rank it from a business perspective very high on the priority list because actual customers were getting actual value out of other things. And it’s kind of like, well, this is just our cross to bear during implementation, right? So I think it was on the plane home from the Growth Elevated Conference this year where I had been messing around with Claude Code. And I feel like despite being an AI skeptic,

Julian Castelli (16:29.756)
I’m ready.

Julian Castelli (16:34.995)
It’s kind of like, well, this is just our cross-deprior of interior presentation. I think it was on the point in front of the Colorado Conference this year. I had been left to cry. I feel like despite being the guy standing, the thing that I love, I believe that on South Central, if Gen Z could probably give any one of their future maps that is left.

John Peebles (16:53.6)
The thing that I love, I believe that when it’s all said and done, agentic coding is probably gonna be one of the very few killer apps that is left. And the reason for that is there’s a forcing function where the LLM generates a bunch of stuff and you have no idea if it’s true or not, if it works or not, but you can execute it with code, right? You can test it. I can’t go out and check if the summary of Tolstoy is accurate or,

Julian Castelli (17:13.795)
You can test it quickly, That’s right. Yeah, either works or it doesn’t, right?

John Peebles (17:21.974)
captures the sentiment exactly, but I can with code, right? And if you think about it, I mean, you and I have been in tech companies for a long time. We’ve spent tens of millions of dollars paying people to write code that we’ve never read, but we go out there and we execute it and we do the demo. The engineering says it’s done and you go out and you’re flying without a net. You go and demo that in front of a customer. So it’s not an unusual experience actually once.

Julian Castelli (17:39.14)
Right.

Julian Castelli (17:48.402)
full experience actually once people were in it. so I think that’s where I think the network of theater is at large. sure that I put this at the beginning of the talk. One thing, because I remember talking from a client to a developer or CCO to Mr. Hick, the public guy, and the went in did a project that was to them, and they said, we can take the link. Please, Roger.

John Peebles (17:49.74)
once you leave the keyboard. so, and I think that’s where having a network of peers that are experimenting with this rapidly is really valuable because I remember talking to Mike, who’s our CTO here at Administrate, accomplished guy, former principal engineer from GitHub, but probably best known as the lead maintainer for the Homebrew project. So almost every engineer that uses a Mac.

ever has used some software that he’s been responsible for. And he and I were always munking around with this stuff. And a lot of it is just trying to one up each other, if I’m being really honest. I’m trying to demonstrate that I’m still an engineer, and he’s trying to demonstrate that I’m actually a suit. And there’s all of this competition things. I remember him telling me in the summer of last year, hey, one thing that’s really cool about Claude is that it’s

Julian Castelli (18:19.819)
That’s healthy.

Julian Castelli (18:43.397)
last year. And one thing that’s really cool about the law is that it’s very good. It’s awesome.

John Peebles (18:49.067)
very good along, know, LLMs are great at summarization and translation and things like that. He said, one thing that’s awesome is you can point it at a minified JavaScript library, I get an obfuscated bit of code and it can basically like explode it out into normal English and code and then it will tell you how it works. And I just, I just couldn’t believe this when he told me and it was like, really? He’s like, yeah, try it. And so I, of course I tried it and you take this.

Julian Castelli (19:12.058)
Translated to something you can understand.

John Peebles (19:17.469)
library that’s been intentionally obfuscated and it will just report the code to stuff you can read and explain it to you and translate it, right? And so I was sitting there on the plane and I probably should have been trying to get some sleep and like honestly, I’d had a couple glasses of wines. It’s not like the clearest like headed moment, right? And it’s on… Yeah, yeah.

Julian Castelli (19:40.562)
So the skeptic in you was vulnerable. You gave it a shot.

John Peebles (19:46.13)
And you’re on plain Wi-Fi. And I just said, OK. And I found the part of our code base. I actually messaged the engineer. I said, hey, where’s all this control panel code? And they said, it’s over here, this nasty area, whatever. Don’t look there. It’s dangerous. It’s whatever. And I was like, OK. And so I just said, hey, Claude, look. We’re trying to build a configuration work

Julian Castelli (20:00.037)
Right.

Dangerous cobwebs.

John Peebles (20:15.531)
tool, workbook, and I found all of our implementation documentation, I found all of our support documentation, I got all of our API information, and I dumped it in a folder and I said, we are going to build an integration tool, or implementation toolkit that basically will explain to a customer, here are the decisions that we need to make and why.

Julian Castelli (20:31.407)
And their visions will that will get that. He’s definitely. Here are the students. Why reference. Our implementation artifacts with all sorts of support documentation. As well.

John Peebles (20:43.179)
and it’s going to reference not only our implementation artifacts, but also our support documentation. I loaded up all of our marketing case studies and said reference these as well. And so every time we say there’s a decision that needs to be made, you will have a reference to a successful customer case study. And then…

Julian Castelli (20:59.629)
An example of significant context.

John Peebles (21:05.203)
So, I said, and then what we’re gonna do is please look at this code over here that’s awful. And people say, you know, it’s got signs up that say beware and don’t tread in this area or whatever. And introspect it, figure out, yeah, figure out how it works. And then, yeah, yeah, exactly. Get in there, figure out how it works, where all the ghosts are and whatever. And then just build out an API that can manipulate it. And it did.

Julian Castelli (21:05.681)
So, I said.

Julian Castelli (21:16.561)
The Code graveyard. The Code haunted house.

Julian Castelli (21:33.628)
It’s it.

John Peebles (21:35.249)
And it was about half an hour, 45 minutes. It almost one-shotted it because again, they’re very good at translating code from one language to another or whatever. And basically we had this script, of scripts that I could run on my local machine. So I didn’t need to deploy anything. I didn’t need to call our engineering leads.

Julian Castelli (21:41.76)
Yeah, right.

Julian Castelli (21:50.037)
Basically, I’m just scripting scripts on a machine, then keeping notes with the way they’re doing it, and making it more easy to read. Because with the power security, it’s a hard task, so I’m quick to do And I did basically tongue in, but I didn’t bring out a bunch of decisions, with all the decisions, for reason. I think that then…

John Peebles (22:01.492)
It was within our secure boundary because our laptops are encrypted and so forth. And I could basically punch a button and it would generate out a bunch of decision, we call them decision trees, in Notion. It would then walk you through these decisions. And there’s a great feature that Notion has, which is you can incorporate little databases within the pages. And so it’s kind of like an access database if you’re old enough to remember that.

Julian Castelli (22:18.105)
Right.

John Peebles (22:28.893)
And so you could fill out things and then there’s referential integrity. So one of the things you have to do is you have to say like, how many companies do you have operating within administrate? And then what regions do you have and what regions to which countries operate in? And because of the way that Notion works, can enforce that they can’t choose a region that doesn’t exist, right? And so you go through all this and then it would just output a YAML file that was human readable. And then you can look at it.

Julian Castelli (22:47.441)
Yeah.

John Peebles (22:56.811)
And then you could punch a button and 30 seconds later, all the config is imported. Your instance is completely set up and that is dozens and dozens of hours of manual knob fiddling and whatever that is now gone and doesn’t need to exist anymore. And the wonderful thing about it is you could then decide, well, I screwed that up. I forgot X, Y, or Z. You can blow the whole thing away and run it again, right?

Julian Castelli (23:17.744)
And then, I screwed that up, and it’s like Y, Z. And we’ll just fix it. just tell the model, change A, B, C, D.

John Peebles (23:30.207)
Yeah, or the customers like, forgot this fine, add it in and we’ll punch the button again. And it’s a dent put in so you can run it as many times and it won’t, it won’t destroy stuff that, you know, it’s already got, imported correctly. And, and that was the first major step that like open our eyes to, wow, there’s these opportunities internally that are relatively low lift that literally seem impossible. Right? If we, if we’d taken that, if I hadn’t known about the

translating minified obfuscated code into that I just somehow heard overheard from Mike and he’d seen it somewhere and and and whatnot then you just would never have attempted a project like that you know how do you do this there’s no API yeah yeah

Julian Castelli (24:07.919)
Right. Yeah, it’s almost too incredible to believe. don’t think, it should be natural. But that’s the key is having an example like that to test out to see what’s possible. And also having a peer to, like you said, you were describing the camaraderie and the challenges that you were challenging each other, right?

John Peebles (24:31.21)
Yeah, exactly. so, you then then you kind of get to the and there were some bugs or whatever here and there, but it was like and it hallucinated a few things on the the explanation stuff or whatever. But, you know, you just I gave it a read through. And then the interesting thing was we ran into the next challenge, which was nobody believed me, you know. So I land. Yeah, I land.

Julian Castelli (24:52.109)
So that’s a human challenge, right? Because it was just too much, right?

John Peebles (24:57.258)
It was, and I mean, I could barely believe it myself. And I’m like, I’m serious. Like I think I have this thing and, and, and, and the second piece of work that I did, which was pretty fun too, is, you know, we use Gong, like a lot of companies to record sales calls. And, so very quickly we built a little script that sucked out all the calls from Gong, all the transcripts, boiled them down and then basically went through and shoved them into a handover document framework.

that then sales didn’t have to sit there and do a bunch of manual effort and and what was really cool about it is I was pretty Adamant about every single time that we had a fact that was referenced it had to be referenced It would be footnoted to reference the gong call and the timestamp Yeah, and it could still do that but at least you can double check it and click right in there and same thing for our contracts it would say

Julian Castelli (25:39.523)
Right, so it’s not making stuff up, so you have that check.

John Peebles (25:52.607)
You know, they’ve bought this piece of the platform and it’s contract page four, line 32, you know. And so, you know, we started iterating on this rapidly. And like I said, nobody believed it. I’ve convinced somebody to look at it. One of our product managers, he couldn’t believe it. And then we just said, hey, like, look, we’ve done this work and we think this will cut out a lot of time. And

Julian Castelli (25:57.943)
Thanks.

John Peebles (26:21.236)
You know, it still took weeks for the team to really kind of understand or get through the shock of, wow, like I don’t, yeah.

Julian Castelli (26:29.479)
Just to digest the new process because they were stuck in the old paradigm.

John Peebles (26:34.078)
Yeah. And so thankfully though, know, they’re, they’re, they’re great. And they, they lashed onto it and we had a couple of new customers kicking off and, you know, it wasn’t like, you know, nothing’s perfect, but we could iterate very rapidly. And again, we’re, we’re at the point where we were doing like on a weekly 30 minute feedback session where, we’re looking for, Hey, this didn’t work very well, or that needs to be tuned up or whatever. It’s.

we’re now at the point where that’s too slow, because we can ship the fixes or the adjustments so fast. And so that’s been really interesting. But one of the wonderful things about it was I actually had the opportunity to meet up on a customer, one of the very first customers that went through this process. And it’s a far cry from we’ve all done

Julian Castelli (27:07.682)
Right, so quickly. Yeah, that’s right.

John Peebles (27:31.688)
a SaaS implementation. We’ve all seen the spreadsheets of Doom or Google Docs, or if they’re really sophisticated, it’ll be Airtable or something like that. But we can provide a login to their section of our notion. They log in, everything’s laid out, it’s explanatory. We still get on the phone with them and explain things, but they walked through this process and they were just commenting. They didn’t know it was a new thing for us. But they were coming out just how slick everything was. And there’s obviously been

Julian Castelli (27:55.906)
Right. Wow, like this is.

John Peebles (28:00.319)
you know, months and months and years of, of refinement and all this stuff. And you’re kind of thinking, wow, like this is, this is wild. And the thing that I love about it. Yeah.

Julian Castelli (28:08.43)
That was the feedback they were giving you on the process and you’re like we just got this together. is brand new.

John Peebles (28:16.232)
Yeah, yeah, it was like, it was like, you know, less than 40 hours of actual wall time, you know, that it, yeah. And it makes our team look like.

Julian Castelli (28:20.482)
Wow, that must have been incredibly satisfying. So you’re reducing time to get the work done, you’re increasing customer satisfaction, you’re accelerating the time for customers to onboard. Those are all key benefits.

John Peebles (28:35.038)
Yeah, and massively improving quality and the perception of quality, I think, as well. It’s also just the customer is coming on board with a much deeper understanding of the decisions that influenced how things work in the platform, which is difficult sometimes to fully internalize. So yeah, think it is…

Julian Castelli (28:49.893)
Yeah.

Yeah, think this is a one.

John Peebles (29:01.746)
that has since grown arms and legs and spawned a whole bunch of subsequent projects that will be familiar to anybody that’s running a similar style of platform business, which is great. The configuration is done, but now we need to move on to that data import challenge. And then there’s other challenges that exist behind that. But I think what’s interesting and what I love about this whole thing is it was just such a

Julian Castelli (29:15.969)
Yeah. Right. I think what interests me the most about this is it’s a nice self-dispute experience. I had to have access to…

John Peebles (29:30.404)
nice self-contained experiment. had to have access to some, our code. I got access to Gong cause I didn’t have that at the time. You know, I got an API key from Claude from Anthropic and, but it meant I could test this hypothesis. And, and it was really, it was a really fun thing to make a lot of progress quickly. And then it didn’t become somebody else’s burden.

Julian Castelli (29:37.826)
Yeah

Yeah, I.

John Peebles (29:59.08)
So myself and the product manager, we’re still the owners of this. he spends probably a couple hours a week on it and I spend 15, 30 minutes a week on it. But it scales, and it doesn’t have to be this thing that we then, the CEO’s nightmare pet project that gets handed off to somebody who actually does all the work.

Julian Castelli (30:00.064)
Yeah. That’s pretty cool. Yeah.

Julian Castelli (30:15.372)
Right, right. That’s terrific. What advice would you have to other tech leaders to kind of cross the chasm, if you will? What was holding you back that you might, other people might identify with and what was your advice be to them?

John Peebles (30:35.411)
Yeah, so I think that to be charitable to myself, which my girlfriend says, and I always am, I think that before Claude code came around, it was difficult to get stuff done and painful, right? Like you could do some coding, but it was not, it’s just ergonomically annoying, right? And the models were not, they’ve improved a ton. So I feel like

Julian Castelli (30:46.868)
I think that.

Julian Castelli (30:51.885)
It’s difficult to get sucked on and painful.

Yeah.

John Peebles (31:05.705)
I don’t think I wasted too much time. I was on it within weeks of it coming out and experimenting and playing and whatever. I think that’s a key piece, which was I started playing on my own hobby projects. Projects making Lego move around and my own personal websites and stuff like that. And I think just that experimenting and, hey, can I try this?

Julian Castelli (31:27.532)
Getting familiar with the tools and what they can do otherwise you don’t even know it’s possible

John Peebles (31:37.196)
Yeah, exactly. then you get it’s like that thing, right? Like we’ve all been I remember going to like this winter camp once and I was like, oh, there’s this snow hill, like snow plaza packed up this huge mountain of snow. Right. And we get out there with like, you know, a garbage can lid or something. you know, guys in college or whatever. it’s like, you know, we we slide down on the garbage can lid. And it’s like, OK, now two of us are going to go. And then it’s.

Julian Castelli (31:51.21)
Right.

Keep iterating, experimenting.

John Peebles (32:02.1)
two guys standing up going backwards, you know, and it’s like, and then I think, you know, somebody breaks an arm and we still keep going, you know, it’s like, it’s like, it’s that type of thing that I think is, yeah, it’s really, exactly. It’s really that iteration and that kind of, cause you start with, can it write me this function? Can it write me this module? Can it write me this bigger module? Can it write me piece of an app? Can it write me a whole thing? You know, that is really important.

Julian Castelli (32:09.612)
You slide into the river.

Julian Castelli (32:21.237)
Yeah.

John Peebles (32:31.859)
to the experimentation is just so, important. And I have personally seen some of the best engineers, some of the people I respect the most that have so much experience in tech, they just do not believe. People will not believe it until they do it themselves. There’s just no convincing anyone. Yeah.

Julian Castelli (32:32.428)
experimentation it’s just a simple film for you. I personally think the best engineer, with respect to most important is the people who are going to leave it. You gotta get your hands dirty, right? I mean this wasn’t something that you could have just delegated easily, right?

John Peebles (32:58.973)
Yeah, and, and, or you could, and when it come back, not possible, or, you know, we tried and that was that. so.

Julian Castelli (33:08.042)
Yeah, well it’s really hard when the CEO programs it and shows that it works. That’s a pretty compelling way to get some change in the organization.

John Peebles (33:16.753)
Yeah, it was funny too, because I was actually, I tried to almost keep it quiet that I had done. So I was like, like we were experimenting. was like, we is like, there was no we. Yeah. And then I was like, we got our product manager. was like, Hesham did this, you know, cause it’s like people listen, you know, yeah. They knew Hesham was involved. Yeah. It’s just like, I didn’t want people to know. then I think.

Julian Castelli (33:27.083)
Wait, the Royal way?

Julian Castelli (33:33.963)
Give some of the credit. Yeah. It was the AI committee.

John Peebles (33:44.125)
kind of word got out and then, you know, it’s kind of like, wow. And it’s been nice to see because now people are experimenting on their own stuff and yeah.

Julian Castelli (33:53.459)
Yeah, that’s what you want. mean, you got to success. You got to celebrate the successes. You got to make it accessible and you got to give people the room to to experiment themselves. But that’s terrific. So so now you’ve got scalability in this area and now more projects are growing laterally. What advice would you have to another CEO who maybe isn’t a, know, as technical as you? What advice would you have to others to just try to implement AI?

in their organizations.

John Peebles (34:25.384)
Yeah, I mean, I think the coding areas is a real sweet spot, right? I remain nervous. know, we took, last week there was an election in Edinburgh here in the UK and my girlfriend and I are going out to vote and I go to the normal precinct. That was where I voted two years ago. It’s not there, right? It’s an elementary school and there’s

there’s a bunch of elementary school kids and people are like, what is this long haired guy doing, like poking around? So I’m like, that must not be the right, I couldn’t find my polling card. I thought I’d taken a picture of it. And so of course somebody pulls out and asks Gemini, their phone and pulls it. Where’s the polling place? it’s down down down many street, just down a quarter mile down there. that’s and I’m like, I just don’t know. Right. Like how?

this training data stopped about six months ago. how, so we look it up and it’s actually, no, it’s the library across the street, right? Like it’s just completely wrong, you know, and never in doubt, never in doubt. And, and the, was on a family trip, my girlfriend and I met my parents and we were out in Italy and there’s, there’s a European view of holidays and vacations. And there’s maybe an American view of, this is our trip across the pond once a year, you know.

Julian Castelli (35:25.138)
But never in doubt like very confidently

Julian Castelli (35:39.079)
Thank you.

John Peebles (35:46.537)
And my parents were taking the American view. My mom lined up 75 stops in nine full days, you know? And they’re all timed entry, right? Because Italy has gone to all timed entry. And so this is very logistically, and actually there’s a whole bit about using AI to do that. But there was like this fresh hell, right? Where part of the trip, the train wasn’t running that day. It had been canceled for whatever reason. And everybody pulls out their own AI and asks it.

Julian Castelli (35:49.81)
Oh, God. Oh,

Julian Castelli (36:01.994)
It feels like this fresh metal, right? Where, uh, part of the trip, the train, that thing, was canceled for. And everybody pulls out, they’re all picking it up.

John Peebles (36:14.569)
And, and all of the AIs are coming back and like, like Gemini says this and oh, know, Claude says that or whatever. And I’m just standing there in the middle of Italy with like, and I’m just like, this is just unbearable. Right? So I remain skeptical about a lot of these applications, but I think the coding area is where you should, you should really have a look. There will be so many areas where there’s simple little jobs.

Julian Castelli (36:23.337)
No, that’s…

Well, I think it’s got to a little bit.

John Peebles (36:44.547)
One just popped up this morning, and it’s a bugbear of ours. It’s been for years. It’s like getting customer certificates like lined up where they want that first name and the last name swapped in on a PDF. And it’s going to be printer ready. sometimes it’s a plastic card. Sometimes it’s a piece of paper and it’s all very important. And that’s a problem that we’re going to tackle because we’ve been doing it manually for a long time. But it’s code. You can validate it. You can see the output.

Julian Castelli (37:02.101)
Yeah. Yeah. I think those are the areas where if the public offices that have just.

John Peebles (37:12.713)
I think those are the areas where these internal processes that will have just grown arms and legs are very difficult, very time consuming, soul destroying. If you can get them automated, it’s low risk. If it blows up, the next person in the chain is going to know. You don’t have to worry.

Julian Castelli (37:18.633)
Yeah. So now you’re seeing new opportunities much easier. It’s it’s almost suddenly they’re emerging from the blur. That’s pretty awesome. Well, John, thank you for sharing that experience with us. Congratulations on that progress and we’re grateful to have you on the show today.

John Peebles (37:35.239)
Yeah. Yeah.

John Peebles (37:44.841)
All right, thank you so much and enjoy the new world.

Julian Castelli (37:50.456)
Absolutely, we will.