George Jagodzinski (00:00):
Today we learned how some of our Intevity team felt 200% more productive with AI, but when you looked at the actual output, the numbers hadn't moved. Figuring out why is where things get interesting. I'm joined by two of our own, Steve Adams, our chief digital officer, and Jeff Wooden, our chief architect here at Intevity. Amongst many other things, Steve leads our client delivery and Jeff is the engineering mind behind a series of launches on Intevity Labs, a set of open source tools born from the real problems we run into serving our clients.
(00:27):
Labs happened organically. We were trying to figure out how to give our engineers real velocity with AI without losing quality, blowing through token budgets, introducing security or compliance concerns, all the real world stuff that makes scalable AI messy. The tools we built to solve that turned out to be useful enough to share and we like to share. So we discussed Sentinel and Trove, the two first open source releases, the nuances and issues organizations run into when they try to measure AI productivity, and how quickly your brain can just fry when overseeing too much AI work all at once. We're also drinking bourbon because craftsmanship matters and I love time with my team. Please welcome Steve and Jeff.
(01:06):
Welcome to Evolving Industry, a no BS podcast about business leaders who are successfully weaving technology into their company's DNA to forge a better path forward. If you're looking to actually move the ball forward rather than spinning around in a tornado of buzzwords, you're in the right place. I'm your host, George Jagodzinski. Jeff and Steve, thanks so much for being here.
Jeff Wooden (01:46):
Great to be here.
Steve Adams (01:47):
Good to see you, George, Jeff.
George Jagodzinski (01:49):
Topic we're going to cover today is AI, Intevity Labs, Bourbon, all sorts of fun stuff. Let's start with Intevity Labs and the why behind it. Why did we create it? What was the impetus? Steve, why don't you go first?
Steve Adams (02:02):
I think the first thing to say is it happened kind of organically. We didn't sit out with the goal to create Intevity Labs. We really tried to tackle two things. One was how do we empower our engineers? So we wanted to give additional velocity. We didn't want to lose quality. In fact, we're hoping to get additional quality out of things and we need to do it safely. Jeff and I talked a lot about this idea of how do we transition from the mindset of traditional engineering to more architecture and orchestration. And then the second piece was really about how are we creating the services in aligned with our client's needs. So we always talk about this idea of avoiding shiny objects and making sure that everything is tied to business value.
(02:41):
So we were getting at, how do we unlock additional efficiencies? How do we get at new sources of revenue or how do we de-risk the organization? And so as we started to go through solving those problems, part of that was creating tooling. And we hit a lot, I think, of the challenges with our clients at scale that a lot of organizations are facing. And so we wanted to share that. I think it was Jeff's idea to say, "Why don't we open source some of this tooling?" And then it happened by nature that things went hand in hand where a tool went with a concept that hopefully aligns with a lot of the challenges that other organizations are having.
Jeff Wooden (03:15):
Align on the same. I think we saw gaps in ways we can make the teams more efficient. And then all the things you mentioned, Steve, but the other thing that comes out of this is the ability to upskill our teams as well by providing a platform that may not necessarily be client work that they can utilize to learn and grow and put out their own applications.
George Jagodzinski (03:39):
We see our clients and people just being inundated with these AI platforms and products, and it's like, how do you even navigate it? What the heck does this stuff even do? And what I love about this is it's just rooted in just getting stuff done. We're encountering problems in the real world and solving with things and not necessarily even turning into a product. It's just like, let's open source it. Let's just help ourselves be better. Let's help our clients be better. To that end, Jeff, why don't you start with just an overview of what are the first couple tools that we're starting with here? And then we'll get into those first and then we'll get into what's next later.
Jeff Wooden (04:16):
The very first tool I released, I think a few weeks ago is called Sentinel. So Sentinel itself is a Claude code companion. So in terms of what I do, I've got my own Claude subscription that's an Integrity subscription. I have a personal one. And depending on what client I'm working on, I might have a client subscription as well. So it helps solve one issue there, which is the ability to easily manage and switch through my accounts. But I think the more powerful features and what we use it for across our teams is number one, security. So even with Claude code, you can set your rules. You can use this bash tool or what have you, and you configure that. But short of that, you have no insight into what it's actually doing. And so when I first implemented Sentinel, security was top of mind and it would show me things that the agent was doing locally, like I'm reading my SSH keys.
(05:11):
And like, well, that's not cool. So I want to know about these things. I want to be able to block certain things if they happen. And so it adds an extra layer of security. But then more importantly, as I look to evolve the tooling, the next major feature that landed was optimization. And so when you think about tokens, they're essentially a commodity and you can burn through them rather quickly. So what Sentinel also does is it offers three tiers of optimization. So one is curated sub-agents. So let's say I'm using Claude code and I'm using the latest Opus 5 model. Well, does Opus 5 really need to read files on my computer? Do I need to pay for that at Opus 5's token cost? Or can I use a cheaper model like Haiku to do that? And so that's what curated sub-agents do, is allow you to install a set of curated sub-agents, see what the potential gains are, and then turn them on to realize those gains.
(06:05):
And then there's another optimization feature for context management. So think of code execution. So anytime you have an MCP server that you're utilizing, all of those instructions are going to the agent every time you use it. And so that's a lot of wasted tokens. Now what the code execution feature does is essentially wrap a CLI interface around the MCP server, allowing you to use batch tooling in order to access that tool versus all of the bloated context when you're trying to leverage your agent. And then lastly, and probably the most powerful optimization feature is compression. So with compression, we can compress a lot of the responses because typically there's a lot of information that the agent won't necessarily need right off the jump when it's getting its response. And so with introspection, it can go back and get the full payload, but in most cases it doesn't need it.
(07:00):
And so this feature alone accounts for over half of token savings in Sentinel. But overall, I find even with my regular usage, I'm typically saving around 70% of my tokens in a given day just by having these optimization features on, which is a huge win.
George Jagodzinski (07:16):
That's huge. Yeah. One of our recent guests in the podcast was Linda Pack, CIO of Estee Lauder. And one of the things top of mind with her was token economics and token governance because it can get out of control fast. And I know, and Steve, you could probably speak to this, but there's a reality on the ground when we're working with our clients and you want to give everyone the tools to the kingdom, but things can get expensive fast. And nevermind everything that you're talking about, Jeff, security and are you using the right tool for the right job and all that? And Steve, what do you see as the reality on the ground?
Steve Adams (07:51):
We've been at this for a while. I was looking back at our first AI project and it was September of '23, and a lot changed since then where, I don't know if Jeff would agree with this, but I feel like when Opus 4, 5 came out, the world kind of changed and suddenly the things that seemed like they were possible suddenly became possible and it's just gotten better since then. We kept hearing the team say, "I can accomplish 200% of what I did before or 300%." And then we look at the team velocity and it hadn't necessarily moved. So we took a step back and we're like, "Why does everyone feel like they're being more productive and we're not getting more out of this?" And I think it boiled down to a couple of things. One is the way people were story pointing had changed because they were starting to story point on what it would take with AI versus the original story points.
(08:34):
So kind of like the systems adjust together. But the second piece, it's bringing this back around to your original question. I got on the phone with Jeff and I was like, we need to be able to have visibility into our token spend because it's not at zero cost, whether it's for our clients or whether it's internal cost and understanding where's the token spend happening? How do we actually understand how that translates into velocity? And so the heavy end of the story is a lot of things had to come together, but we hit this inflection point where things just went vertical in terms of what the teams were actually able to accomplish. Part of that was getting the newer models. Part of it is the systems around the models, the harnesses and our process and things like that and being able to see that.
(09:15):
So just to ground it in a concrete example, we have one client that was, we gained nine weeks on the schedule over a period of about between two and three weeks, just from the point where we hit that curve where we threw all the tools that Jeff has pulled together, we got the process in place and everything just started to click. And so I think a lot of organizations from hearing is like they're still waiting for that to happen or they're maybe feeling that they're doing a spend, but they're not seeing the return yet. So a lot of things do need to come together, but we're really starting to see that and click on all the cylinders.
George Jagodzinski (09:46):
I love that. I want to poke into the story pointing thing because it's important, but I'm going to pause because I think we're also... The craftsmanship is coming through here as a team and I think we're going to have some brown liquor here, which represents craftsmanship. And so why don't we go around the horn, let's talk about what you guys chose and why. So Jeff, what do you have there?
Jeff Wooden (10:07):
Yeah, so I grabbed the Makers number 46. I think one of the first bottles of bourbon ever gifted to me was Makers. And I saw this one and I think seven of the stabs of the barrel were made with French Oak, which is a little bit tighter green, maybe a little bit more tannin. So this will be my first try. Excited to see what the French oak offers.
George Jagodzinski (10:26):
Nice. And Steve?
Steve Adams (10:28):
I have Chris Stapleton's whiskey, so, Traveler. And that was partially because I have a concert coming up in a week and I'm excited about it. And I was in the liquor store and I was like, I'm going to give that a try.
George Jagodzinski (10:38):
It's also on brand with one of our favorite clients, Sazerac.
Steve Adams (10:42):
That's a good point. And I didn't realize it until I got it home and actually read the label that it was a Harlan Wheatley collab. So that's really cool.
George Jagodzinski (10:50):
Well, I went with Weller 12, which is also on brand with our client here. What I love about Weller is Pappy Van Winkle gets all of the street cred and the brand recognition and everyone think that's the best, but Weller 12, it's the same liquid in the same barrel essentially. And so you're paying for the craft rather than the name, which I'm always so big fan of. Just get the good stuff done. We don't need to be too braggy about it.
Steve Adams (11:15):
That's why they call it Baby Pappy, right?
George Jagodzinski (11:17):
Yeah, that's right. But back to the story point thing. So for those not familiar, story points are how you estimate the level of effort for any given code that you're going to develop or functionality that you're going to deliver. Actually, we haven't had this conversation yet, Steve, but are we at our clients, are some of our clients looking? Are we comparing what it would've been before versus AI versus non-AI? Because I feel like a lot of organizations need to track both of those so that when you get the big bill at the end of the month for all your AI spend, you could compare and contrast the ROI there.
Steve Adams (11:46):
We definitely are. And that was kind of what I was getting at, that there was a leg in actually showing the progress from what we're hearing. So hearing from the team and everything's going great. People are really making good use of these tools. And I didn't question it, but I was also wondering why aren't we seeing that? To your point, some of it is how we're measuring it, but the other piece is just getting all the pieces to fit together. We have really increased the velocity that is going through in the sprint in the last few months.
George Jagodzinski (12:15):
That's awesome. And Jeff, well, both of you, but Jeff, I'll start with you. I know you're super busy. We're all busy. Everyone's busy. And so you found time to really dive into this rabbit hole and start building these tools to create productivity. I think a lot of people, they're sitting around and they're saying, "Man, I don't have the time to do this. Someone needs to carve off some time for me to do this. And maybe I need a week and I could do it." Talk me through a little bit of your personal journey as to why you started doing it and how you fit it in around your regular responsibilities.
Jeff Wooden (12:51):
When my journey started, was it maybe back 2021 when the ChatGPT first launched? It was kind of when I started diving into how can we leverage LLMs for engineering? And that was a long road. It didn't change things significantly. You're still spending a lot of your own time reviewing the work that it was doing. And a lot of it was just bits and pieces. Now fast-forward to end of last year, beginning this year when the latest frontier models dropped and everything just changed from there. And I quickly realized, well, a lot of this is about duplicating my efforts. It's not necessarily cloning myself, but working concurrently with AI. So while I may be able to work on one client project, I can work on a labs project concurrently, start my prompts, my planning process, and then iterate that way. Now there is still, I would say, no matter what project you're working on, it's like that last stretch, that last 20% of the project is where all the time is spent.
(13:56):
So I can do all the planning upfront I want, have my agents implementing it. But at the end of the day, I've got to put in the real QA work, make sure things are working, matching my intent based on what I wanted to build. I find that my day-to-day now is typically at least minimum five concurrent workloads. And then I'll find the time to focus on that last leg of the journey for a particular workload. And that's not to say that things aren't continuing to change because they are. While we may still do some manual prompting, depending on the client and what gates that they have, the future is really loop engineering, and that's going to help change things even more drastically beyond just the workloads I can manage as a human when they're all running somewhere in the cloud and things are just working and you're validating based off of intent and goal. So we'll continue to see it evolve, but that is my typical day today.
George Jagodzinski (14:50):
How about you, Steve?
Steve Adams (14:51):
I want to key in on a couple of things that Jeff said that were kind of nuanced. One is we talked about the team finding a rhythm and getting to the point where we have the process, the tools and everything else in place so that they can be productive. One of those things we kind of skim past is this idea of we have session limits in terms of we talk a lot about tokens upfront and you actually need to sit at the beginning of your day and say, "Okay, I'm going to plan out my day so that I can use the tokens appropriately so that I'm hitting a refresh window." And things like that. Then there's other tools that have come on board like remote control that everybody wants to step away and manage your personal life and your professional life and things like that.
(15:32):
But I know Jeff has talked a little bit about this, that you have this basically whole team ready to go. And from now, from your phone, you can say, "Okay, now do this, now do that." So you can be out somewhere and it's coming back to you proactively instead of having to basically be at your computer all the time. And what that does, it's kind of like a double-edged sword because you get a lot of power with that, but it's also hard to detach. So when you said to Jeff, I mean, Jeff has really leaned into this stuff. So I think it's been a challenge to say, "Okay, how are you not always on?" And things like that because there's a lot of power with it. The other thing I wanted to note is it occurred to me with what Jeff was saying is there's kind of two things that we unlock with AI.
(16:12):
And one of is this idea that we can automate the simple task. So Jeff mentioned this idea of, okay, there's a whole bunch I do every day that I don't want to be doing, whether it's orchestrating the test cases or some of the other more in the details versus the architecture. So there's a ton of value in that just in terms of the efficiency. But there's also this thing where it's like, I can now unlock something that wasn't possible before. And so that's where the transformation side comes in. I think Jeff does a great job of doing both because I hear people talking about this idea of it's one or the other. And it's kind of like if you can automate the simple things, then you can open up the time to do the things that weren't even possible before.
George Jagodzinski (16:51):
Yeah, totally agree. I don't know about you guys, but I've gone through this rollercoaster of, I loved that part, the double-edged sword, you're so right, Steve. Because I'm like, oh, when I'm waiting in between meetings, I can now create this thing that would've taken me three months. That's awesome. But now I'm finding this mental load of... At first I thought it was great because I've got total ADD and I got a computer science background and I've been around for 26, 27 years. So I got the pattern recognition and I know how to delegate tasks. And so I was like, this is my superpower. I was built for this.
(17:24):
But then I feel like a few months in, I was like, I'm frying my brain because it's just such high level kind of decisions and thinking. And I'll be honest with you, I haven't figured out how to balance it yet because it reminds me of my early coding days where you have to go to the bathroom or you're waiting to go get a coffee, but you got to finish whatever you're doing on your build because you know it's going to take 10 minutes to run. And so you're just always in this mode of like, I need to go, I need to go, I need to go. Have you guys figured out how to balance that yet?
Steve Adams (17:53):
I'll say what Jeff said a second ago, he's got five sessions going at once. That's a lot. I was kind of in figuring out if I have legit projects going, especially not connected with each other. The mental energy of shifting back and forth. I have four screens here. So it's like I could run three projects at once and then my brain just got fried and you end up at the end of the day just feeling absolutely spent. And so Jeff says he's doing five, and I don't doubt that for a second, but it's a hard thing to do.
Jeff Wooden (18:21):
Yeah, five's definitely a lot. And I think I was reading Boris Cherny, the creator of Claude code. He does something towards 10, right? Although they're geared more towards loop engineering now. But to me, the real load, again, it comes in the last steps of getting something out. It's like that last 20%, that is real time and you got to figure out when you're going to have the time to take that time. And that is a challenging part. Having five concurrent workloads, agents working, doing their thing, that to me is easy, but that heads down time to validate what they did, ensure it matches your intent. That's where the real work is. And I think those are the areas where I'm always looking at how can we improve the validation stages so they can go much quicker and we can get the product out much faster. Still don't have the answers there.
George Jagodzinski (19:12):
I guess one thing I've started doing, I don't know if it's going to work yet, but just blocks of times or specific days where I'm like, all right, on this day or these blocks of times, I'm just giving insane amounts of prompts to all these agents and just sending them crazy and iterating, iterating, iterating. And then I'll maybe block off a day where I say, okay, this isn't about iterating. It's about me that's looking at everything and making sense of it. And how do I add the human layer on top of it and dedicating the... And there might need to be a hike involved in between those days where I could just clear my mind and do it.
Steve Adams (19:45):
This gets one thing I'm thinking about too that we're seeing with a lot of our clients is these are large scale projects with tens of engineers on them in some cases, and they're generating way more code than they were. So with that productivity becomes this huge amount of code that humans aren't necessarily touching firsthand. So they have to go through a code review process. And then to a certain extent, there's this black box of code that nobody's ever touched. And so that's a lot of burden on the people who are doing the code reviews. There's also some risk with that. So we call that dark code. And it's basically this idea that without having hands-on, there's this inherent risk that there's code that's potentially going live or that's in products that nobody's ever touched. And it's not until something goes wrong that even people realize there's a challenge. So I think Jeff, you were dancing around that when you were talking about the fact that all this code is being generated. Now you got to step back and say, "All right, should this go live?"
Jeff Wooden (20:39):
Yeah, that's very true. And I think when it comes to our labs work, this is where we're able to experiment more with how we deal with dark code, how we validate it. We still have clients though who want to have human gates. They haven't moved past the pull request review. So those are very real things. And for those clients, we try to instrument tools that, well, let's have an agent do the initial PR review and then they can give that information to the engineer so we can speed up those human gaits. But it's a very real concern. When I'm working on labs projects, my agents might write thousands of lines of code in a day. And just as a human, I do not have the bandwidth to review it all line by line. So I rely on other techniques to validate beyond just the PR reviewer reading the actual code.
George Jagodzinski (21:32):
I'm always starting to compare it to how different is it from the other patterns that we've had? Because just going back to the before times, you'd have a bunch of junior engineers and you'd have to review the code and you're like, Steve, to your point of dark code, you could insert someone's name there and be like, that's the code Jimmy wrote. Who knows what the Jimmy code could look like? Or the times you'd have maybe a senior person who's really talented and they'd come in overnight and rewrite everyone. They'd refactor the whole code base, rewrite everyone's code base because they think they know better. And everyone comes in all pissed off and then you have to balance that out. And I feel like a lot of those patterns are the same. I'm trying to figure out what's different.
Steve Adams (22:13):
No, I totally agree. I think AI exposes a lot of the challenges we've had for the last 20, 25 years. So the thing that actually popped into my head when Jeff was talking was Gartner pace layering where you've got your systems of innovation, your systems of differentiation and your systems of record. You have to treat them differently, the teams that work on them, the rigor that you put around them. And so we're talking about AI and how much you automated, how much is manual, how much is human in the loop, human on the loop, which means basically AI is doing the bulk of it, but you've got somebody monitoring, observing if something goes wrong or fully automated. And I think there's this idea that you move to the right on that scale to fully automation and that's winning, but it's not necessarily if the stuff that you're working on is really critical.
(22:58):
If you're working on patient data or something like that or some sort of system of record, then it's really important that you're slowing down and you're actually having the human in the loop. So again, same problem to your point. It's just making sure that you kind of restate what success and what winning looks like isn't necessarily just automating everything.
George Jagodzinski (23:17):
Yeah, totally. So Jeff, let's talk a little bit about some of the other stuff going on with labs.
Jeff Wooden (23:22):
Yeah, so we talked about Sentinel. Our follow-up release, which we dropped last week is called Trove. Essentially what Trove is a OTEL collector for all of your AI agents. So whether you're using Claude code, OpenAI, Aiter, OpenCode, dozens of harnesses that are out there. And what we found why we built it is we don't just use Claude code. We have other agents that we have access to, whether for adversarial purposes or what have you. Or we might have clients that say, "Hey, this is our agent ecosystem. You're only allowed to use OpenAI models or what have you." And a lot of these agents also do not have Native OTel being Open Telemetry. That's the observability standard in this day and age. And even for tools that do have it, let's say maybe I have a cursor subscription or Claude code through Anthropic, while both of those organizations have their own analytics dashboards, they're completely separate from each other.
(24:22):
So if I'm on a team and I've got access to Claude code, maybe I'm also using Anti-Gravity CLI, and maybe I'm using OpenAI as well, I want to be able to collect all of that information in one place. Especially as an engineering lead, you might want to see how your team is using the various tools, what the token spend looks like. You could even use that data to make recommendations. You don't need to install this MCP server globally, use it at the project level, things of that nature. But it shifts beyond engineering as well.
(24:55):
As somebody in finance, I may want to understand dead seats. So if I've got Trove rolled out across all of engineering and maybe I'm a larger organization, I've got a thousand seats, but only 200 are being used, that's a lot of waste that we can cut. And so Trove helps to identify that and then make sure that no matter what tool you're using, you can collect all of your metrics in telemetry in one single platform. And that would be whatever your observability platform of choice is. So OTel being an open standard that pretty much everyone supports today.
George Jagodzinski (25:27):
And something like you combine Sentinel and Trove, the multi-agent aspect of it I think is important because a lot of organizations are trying to figure out who are we going to make our gamble on? And a lot of them are going with one right now because it's complicated to manage multiple. But we've experienced and everyone's experienced anthropic will go down, OpenAI will go down, and you're just sitting there twiddling your thumbs saying, "Oh man, I had this really productive day planned out." And by the way, that day was like a 10X day because we're using AI. And so now you've got 10X days that are gone in a span of a few hours. And are you really going to go back to doing it the old way? So I'm curious how you're seeing that play out with those platforms.
Jeff Wooden (26:12):
Well, yeah, I mean going back to the old way is untenable when one of these goes out, and that's something I've definitely experienced firsthand. I think there are certain tasks where if a model goes down, it is a bummer because you really needed that model or you preferred that model. But we have backups that we can use in some of those cases. But I would say overall, the different models we want to use are for certain purposes, primarily adversarial purposes. So for instance, if I am leveraging an anthropic model and I'm putting some work out there and I want to review it, do I use the same model to review the code? It's like having me review my own code. And so that's where another model might come into play and we might do some adversarial reviews to see what that might catch versus the model that actually did the synthesis or implementation.
(27:05):
So all of that's important. But at the end of the day, it is rather a bummer when one does go down and you've got to find a complimentary one. So if Anthropic goes down, you'll probably shift over to OpenAI models. Although there are some others that are catching up, but certainly not at the same level in the benchmarks today.
Steve Adams (27:24):
One of the things we've talked about is Jeff and I talk at least once a week, probably more like once a day. And we're saying we feel like we move in a fast-moving space in general, but we talk every week about what the sort of stuff we've seen having news in AI and it's like one week to the next, it's wildly different. So what that model is and things like that.
George Jagodzinski (27:42):
No, yeah. I was going to ask with our clients, I feel like so many of them are committed to one and they're trying to figure out what's next. And I feel like we're at the forefront of this, but then I also at the same time feel like we're way behind. I always feel like we're behind and it's so hard to feel like you're keeping up. And maybe I just want to say that out loud because for anyone listening, I would assume feels exactly the same way. And I don't know if you guys feel... Do you ever feel like you're on top of it?
Jeff Wooden (28:13):
No. I think it changes so quickly. It is really an effort to try to keep up with it. And I find the more I get used to using agents, I also use them to help me keep up with it. Collect the latest information for me and give me my daily briefing on what's new with AI. So things like that to help me keep up to speed. But while we're even on this chat here, something new is probably occurring with AI that we'll need to know about. That's how fast things are changing.
Steve Adams (28:43):
One of the things that occurred to me, this is a slight jump in topic, but certainly related. When we talk about keeping up with things, one of the things that I've found is a lot of the AI projects are really data projects at their core. And so we've thrown on the term MCP a few times, but basically we think of that as your USBC drive for your data that we're throwing in front of data sources. And we're finding that in order to allow your AI to leverage, to be effective in the organization and truly get whatever the goal is, a lot of times before that there's a data project where we're saying, okay, how do we expose that data? How do we expose the right data? And there can be a project in itself because it's a classic case of garbage in, garbage out.
George Jagodzinski (29:28):
Yeah. And it's not just about the data, there's legal involved, there's compliance involved, there's security involved. And you talk to the head of cyber at any organization, and this is the stuff that scares them the most is what is my team doing with this? I'm sure they could do interesting things, but are they about to expose the largest attack vector to our organization or data leak that we've ever had?
Steve Adams (29:51):
Right. And that goes hand in hand with the fact that it's like for years, those policies in terms of what does it mean to operate in the cloud? Those are established with this At this point, unfortunately I've been doing this long enough where there were conversations around, we can't do anything in the cloud. Everything needs to be at Rackspace. But those conversations a lot of times haven't happened. The policies don't exist. So we're going in and having bigger conversations about what's possible, but it's slow because it needs to be determined like, okay, what's actually allowed?
George Jagodzinski (30:22):
Yeah. So let's give a little teaser on what's next for Intevity Labs.
Jeff Wooden (30:27):
Yeah. So in terms of what's next, we do have a small release coming up called Witness. So essentially what Witness is, if you're typically used to using your agent through a CLI, which most engineers are typically leveraging CLI, some may use the Claude desktop app or whatever other apps out there for the models or tooling that they're leveraging. But what Witness does, and for me, I'm a Claude code user, so when I'm putting together a plan with one of my agents, I've got to read that in my terminal. And sometimes that can be tough. For me, I'm a dark mode user, so it's white on black text. Yeah, it does some little highlighting here and there, but that's not as efficient. And I believe it was OpenAI or maybe it was Anthropic. One of them put out a study about increasing the efficiency of plans by making them more human-readable so you can get to them quicker and understand them before you approve them.
(31:22):
And so what Witness does is it takes the same plan that Claude code would create and it opens it up in a browser in a human-readable format. So it'll generate charts, tables, tons of different assets to help make it more legible for the human, get to the information you need so you can quickly understand the plan, adjust the plan or approve the plan. And so that's the idea is just to kind of speed up the planning process. It'll be released as an NPM module. So again, a smaller release for us, but that's Witness. And the more exciting one right now, the name we've given to it is Curtin. So as I was building Sentinel and Trove, we also wanted to put together some GitHub sites so we could have some documentation for our users and some more information about what these amps were.
(32:10):
I was looking to record some demo videos on usage for these applications and I though to myself, well, this is probably going to take me 20 to 30 hours to script these, manually record them and step through all of that. I was like, why can't I just have AI do that? So I kind of explored a little bit and realized we can create videos like that programmatically. So what Curtin does is it'll evaluate your application. It works with any web-based property. So whether that's something that you compile to a desktop application using electronitory or something that just runs in the browser, it'll script for you and you can adjust those scripts as necessary. It'll seed test data for you, and then it'll step through and handle all of the recording for you. And what Curtin brings to that process to assist with it is essentially a virtual desktop, virtual browsers, virtual CLIs, so it looks real as it puts your application within this virtual environment to record.
(33:11):
And so that brought that potential 30 hours just for Sentinel, which is where I started, down to just an hour after I built this and it was able to record over 15 demo videos that looked great. So definitely excited to get that one out. We use it on Trove as well. Right now while it does work and it's not released, the thing that we're looking to get out along with that is a companion desktop application. So that's essentially the holdup on release is finishing that up. And what that does is allow you to see the videos in real time that AI is making. You can annotate frames, tell it, no, adjust this, adjust that. And then that just goes right back to the agent so it can update your video. So pretty excited about that one. I don't think there's anything quite like it yet out there on the market. So hopefully we'll get some usage out of it once we open source it.
George Jagodzinski (33:57):
Very exciting. And I always love that moment where you start, even you who are really pushing limits with this stuff is you're starting to do it the old way and you're like, why the heck am I not doing this with AI? Why am I not doing with AI? And let me just try moment are the really funny moments. And then on the witness side, I mean, old guy here, but what that reminds me of is command line interface, like the matrix when they're like, oh, it's the woman in the red dress. Where if you can make the woman in the red dress easier to see or when you used to do builds and you could see from a pattern of the text just flowing up the screen, oh, there was just an error in the build. And I know what that error is. And not reading the text, but you could just see the pattern of it. That's really cool to see that coming around. Steve, what are you excited about for what's happening next in Intevity Labs?
Steve Adams (34:45):
We're trying to keep on a pace where we're doing two blog posts a week and we're trying to do a pattern where it's kind of like something on the tooling side and then something related to more of the topics that are related to process or how people are applying to their business and stuff like that. So we've got a couple posts coming up related to the maturity curve in an organization in terms of where you are in this curve and how you can start to plan around it. And then I mentioned this a little bit earlier, but the idea of the autonomous stages and how those two work together, noting that there are places where you want to slow down and places where you want to automate and speed up and tying that to the places where you are in your organization and compliance and everything else we've talked about here. So seeing those things come together at the macro scale I think is really interesting and one of the things that's moved a lot in the last couple of months.
George Jagodzinski (35:35):
I'm super inspired by what you guys are doing and the entire broader Intevity team that makes this all possible. I'd like to finish on what's the best advice that you've ever received in life or work? Start with you, Jeff.
Jeff Wooden (35:49):
Oh, best advice. It's always work hard, stay humble, be kind. That's typically how I try to operate my day to day. But if I harken back to probably one of my favorite quotes, and it's a long one, it's the man in the arena, if you guys are familiar with that one. And so to me, what that speaks to is try no matter whether or not you succeed. It's putting in the effort and saying that you did that. I always think when it's the end of my time, I want to be able to say, "I climbed that mountain." Not that, "I could have climbed that mountain." So if you're not familiar with the man in the arena, it is a Theodore Roosevelt quote, a really good one that you guys should look at.
George Jagodzinski (36:29):
I love that so much. How about you, Steve?
Steve Adams (36:32):
It's kind of funny, we're talking about AI and the fact that how it can work its way into your life and it's hard to shut off sometimes. It also can give back some of that balance. So it was funny if we can use these tools to give us some balance back, I think that'd be great too.
George Jagodzinski (36:50):
I love it. Both of you guys bring it back to the human aspect. Thanks so much for being here. This is awesome. Love the work you're doing.
Steve Adams (36:56):
Thank you.
Jeff Wooden (36:57):
Thanks, George.
George Jagodzinski (37:00):
Thanks for listening to Evolving Industry. For more, subscribe and follow us on your favorite podcast platform. And pretty pleased, drop us a review. We'd really appreciate it. If you're watching or listening on YouTube, hit that subscribe button and smash the bell button for notifications. If you know someone who's pushing the limits to evolve their business, reach out to the show at evolvingindustry@intevity.com. Reach out to me, George Jagodzinski on LinkedIn. I love speaking with people, getting the hard work done. The business environment's always changing and you're either keeping up or going extinct. We'll catch you next time, and until then, keep evolving.