Subscribe
Share
Share
Embed
Welcome to Chuck Yates Got A Job with Chuck Yates. You've now found your dysfunctional life coach, the Investor Formerly known as Prominent Businessman Chuck Yates. What's not to learn from the self-proclaimed Galactic Viceroy, who was publicly canned from a prominent private equity firm, has had enough therapy to quote Brene Brown chapter and verse and spends most days embarrassing himself on Energy Finance Twitter as @Nimblephatty.
00;00;01;05 - 00;00;35;09
Unknown
Slow to do what he feels like. I will do this very fast for this work. Thank you. Pay in dreams and, take my name. John. Slow. Okay. I'm going to profess and be the dummy that I am. They tell me you're like our guru of a gigantic orchestration. What does that even mean? Yeah, I figure I have enough of my net worth tied up in.
00;00;35;11 - 00;01;01;04
Unknown
You know, I don't figure out what that means. Yeah. For sure. So, as I is being adopted in the world, a lot through consumer apps like ChatGPT and, and claw, which is accessible through any browser or app or application on your phone. People are using it to communicate, ask questions. Use it in lieu of like, Google search or something like that.
00;01;01;07 - 00;01;31;17
Unknown
Well, the dev world has found is that it can help generate code on the fly and, generate the ideas that you have without you having to manually code and type and do all this stuff. And, my journey with that kind of started, you know, back in 2023, where I started using ChatGPT to help generate code snippets of ideas that I had and do things that were tedious, like, writing out CLI commands or write out a function.
00;01;31;19 - 00;01;53;22
Unknown
But I kind of got in that mode where I found myself copying and pasting from ChatGPT back to my idea, kind of like, we did with Stack Overflow and, nowadays, we're trying to automate that process. We're not doing copy and paste anymore. We're using smart dev tools like Cloud Code, to generate our platform.
00;01;53;24 - 00;02;22;22
Unknown
And, but what I'm doing is I'm building an agent hierarchy, that's able to, delegate tasks, use tools, and work in a really dense environment like we have in the oil and gas industry. So maybe the best way for me to understand this is we hired you. What did Callide look like then? Versus what does it look like now?
00;02;22;28 - 00;02;53;05
Unknown
Yeah. That's a great question. So I've done this before. Yeah. When I showed up, I think there was, a great vision that I could be kind of the glue in the connector between all these legacy systems that exist, in the industry space. It's it's an industry that builds a ton of integrations into, ERP systems, tier one systems in the enterprise.
00;02;53;05 - 00;03;34;28
Unknown
And they all, have connections to stuff like, well, have you or Aries or, Azari and trying to embed something like AI into each of those in use. AI in each of those applications is really going to be, a harrowing experience. And I felt like Clive was going to be that umbrella, that be an entry point to communicate with all of that and derive answers, from a holistic standpoint, because, you know, the I mean, the brilliance of software, you know, the advent of software, you're too young, you don't remember this, but the brilliance was it actually separated the enterprise into functionality.
00;03;34;28 - 00;04;24;08
Unknown
Right. Let's do customer relations CRM stuff over here in Salesforce. Let's have databases here with Oracle, etc. and it broke up the functionality. And I think AI's brilliance is going to be tying that all back together. Yeah, exactly. Yeah. So when I showed up, Callide was, heavily invested in RAC search. So being able to pull documents across those different domains and derive answers, where I saw the vision and, being able to basically embed an agent hierarchy into the Callide platform, would not only enhance the ability to read from all those systems and documents that, customers have in their enterprise, but also write back to the systems.
00;04;24;10 - 00;04;56;16
Unknown
So automating those integrations, instead of having to build them, piece by piece collides, is going to really be the glue that you can set on top of, all those legacy systems. What are what are some of those agents and like, why would an agent be higher in the hierarchy versus another agent? Yeah. So when people ask that, I think of the user experience in something like ChatGPT or Claude or perplexity, when you ask a question, it is super fast.
00;04;56;16 - 00;05;19;00
Unknown
The way it responds, it's doing web searches in parallel. And, the response time is, is it's just constantly improving. It's always super fast when you try to do something like that in a dense environment, like an oil and gas enterprise, you're going to have, a lot of waiting time if you're just calling all those different systems sequentially.
00;05;19;02 - 00;05;45;27
Unknown
So what we're doing with agents is basically delegating tasks, across the board to go search every data source. That could be helpful in answering a question. And, you have to do that in a deterministic way, even if we're dealing with, non-deterministic models. So the hierarchy, basically consists of a route agent that can delegate tasks to sub agents.
00;05;45;27 - 00;06;10;29
Unknown
And those sub agents have tools which are basically, connections to the systems that we're talking about. And not only are they able to read from the systems, but they're able to write back, using the identities of the users who are making those queries is maybe structuring the Roman Empire. The good analogy here, it's, it's a complex problem.
00;06;10;29 - 00;06;44;04
Unknown
And, I think one of the issues right now is that there's so much noise out there with AI and influencers and being the next, company creating the shiny toy object. But the reality is, it's really hard getting some of these prototypes and POCs into a production environment, not only to work like, you know, for a single customer, but multiple customers working, in a multi-tenant SAS application or a self-hosted deployment, making that all work and configurable.
00;06;44;07 - 00;07;06;19
Unknown
That's a big challenge that we're tackling here at Kellogg. Yeah. Because, you know, one of the things I've, I've thought is going to be really interesting as we go through this journey with our clients, is they're almost like two vectors on this. One is there are times when I want to search and I only want the answer from my data.
00;07;06;21 - 00;07;26;11
Unknown
There are times I want to search and I want to know my data, but then I also want to go out to a larger subset of data, maybe best practices on something. How do I shut down a well, here's my operating procedure, but I'd like to know how Daemon's unit or continental or anybody else, you know, what's best practices.
00;07;26;13 - 00;07;47;08
Unknown
And then there are also going to be times where I literally want to go to the web, you know, I want every bit of information possible. And then the other kind of vector laying over that is there certain times where I need the exact answer. What was the pressure on stage 57 of that? Frak, I can't get the wrong number there.
00;07;47;08 - 00;08;18;17
Unknown
I need to know that there are other times where, you know, I'm looking for a saltwater, disposal. Well, conversion candidate, I need casing probably bigger than five inches, but maybe. Maybe it can't be bigger than, you know, nine and 5/8 or, you know, whatever. I just kind of need to be in that general direction. And then, you know, there are other times where I literally want to suspend belief and ask for an answer.
00;08;18;17 - 00;08;40;01
Unknown
I mean, because if you think about the shale revolution, literally some dude said, let's put baby powder into a sauce rock, right? You know, and everybody get you got laughed out of the room. And lo and behold, that that worked. And so I think that's going to be kind of those two vectors trying to figure out how we do it.
00;08;40;01 - 00;09;03;24
Unknown
And is it something Clyde does? Is it something the client directs and we tick on different databases? I think that's going to be like one of the most fascinating things we do. Yeah. Agreed. And that's, that's kind of where I was getting with, the dance environment that I mentioned because, you know, the consumer apps, they're all just pulling web searches and publicly available data.
00;09;03;26 - 00;09;32;05
Unknown
But those models aren't trained on private or proprietary data sources that are, you know, hold within an enterprise. So being able to ask a question and depend on accuracy, it really, is dependent on an exhaustive search that's going to look for every possible data source and, make sure that we can, you know, tie that answer back to a document or something concrete.
00;09;32;07 - 00;09;56;22
Unknown
Yeah. So our search is we'll bring back citations. They're auditable. We're able to, run queries and configure multiple databases and show how the answer was derived, not just, you know, believe in the Blackhawks right now. That's exactly right. So give me. How long have you been here? Just over five months to seven. Five months.
00;09;56;22 - 00;10;24;15
Unknown
Give me, like, one thing you're really proud of. You want to spike the football, something you built or whatever? Oh, man. So what I'm really proud of is, honestly, this a genetic orchestration system that we're building. There's a lot of frameworks, and, you know, startups that are touting their, their agent abilities or the ability to build on their platform.
00;10;24;18 - 00;10;55;25
Unknown
But a lot of the times when you're building new agents, that requires a code deployment and for the extensive agent hierarchy that we're building for each of these customers, that they can either, you know, host with us or host within their own, environment. They need to be able to, you know, make configurations on the flies or use our PhDs for deployed engineers, to make changes on their behalf, or maybe, add new features, and we're able to do that without code deployments.
00;10;55;25 - 00;11;21;22
Unknown
Those those changes, are evident in the agent right away. So if I wanted to add, a web search tool, there's no version upgrade or deployment. It just works. That's something I'm like, really stoked about. It makes this cool. The hierarchy super flexible. So what's something cool Clyde's going to be doing within the next 12 months?
00;11;21;22 - 00;11;44;29
Unknown
That's kind of going to blow your mind. Yeah, I'm starting to see this right now with the projects I'm working on with, some of the PhDs. We're bringing in, a mapping interface into, you know, our primary, platform being able to render all the, GeoJSON data with layers and interact with it on a chat basis is pretty incredible.
00;11;44;29 - 00;12;13;18
Unknown
I mean, the bi directional nature of that is so cool because historically, you had a map on a sheet of, on a sheet of paper on the table, right? Yeah. And you looked at that map and you kind of had to mentally peel things off it if you want it or whatever. Just being able to chat, pull up on my walls, get rid of all the wells ten years old or, you know, do only wells in these formations add this, you know, blah blah blah.
00;12;13;19 - 00;12;35;08
Unknown
Okay, map that for me. Yeah. Or start with every bit of my data on there and just start extracting things off it. Take these wells off. Take that well off. Yeah. Add this drilling program whatever that it's it's cool. It's a big unlock. It's, because you don't have to have a GIS department anymore. Okay. Go build these maps for me.
00;12;35;14 - 00;12;58;17
Unknown
Yeah. The the interesting thing is that we're able to still ingest and use, the different layers that you'd find in a typical GIS program. It's just the user experience, I think is enhanced with the chat capability and being able to interact with the map like we have, and super fluid. I, I'm so blown away by what are these are creating did, did you hear what I did?
00;12;58;17 - 00;13;29;08
Unknown
The Canisius. So you know, Canisius is day three at, collide. And for background, Canisius is our CTO. He's a software guy, not an oil and gas guy. So we were at, at the client, big conference room table, and I told him, hey, our mapping interface, whenever we build it, the geologists will actually want to see it upside down on the screen because they're used to being at the top of the table, looking at the map upside down.
00;13;29;10 - 00;13;50;19
Unknown
And, clinicians believe me, for about two weeks and, Canisius and Collin are in the conference room and I hear Colin go, jock, do not tell Canisius that maps are going to be upside down the screen. There is. Did you forget about it? Yeah, I totally forgot about it. That was so great. The, So what keeps you up at night?
00;13;50;21 - 00;14;13;21
Unknown
Like, what's the what's the scary thing? Man, I, I wouldn't say it's the scary thing is your 27 kids, but, Yeah, my, I love my kids. I'm blessed with so many of them. But, frankly, enough, on that same note, night as well, I have quiet. So all my thoughts start to come alive, and, I collide.
00;14;13;21 - 00;14;44;26
Unknown
I've been, you know, waking up with thoughts at four in the morning. And I'm just so driven to go, solve all these challenges, production izing, hardening the platform in a manner that we could deploy it in, a customer environment or a SAS environment. There's a lot that goes into that. And then, you know, thinking about how we could do more with less with all these dev tools that are available today, we're able to work, at an accelerated pace that's unprecedented.
00;14;44;28 - 00;15;13;23
Unknown
And with that comes, you know, learning guardrails, harnesses and being able to keep up with, you know, the rate at which people are able to produce code. We're we're in a position where we have to, you know, take ideas that can be, you know, built in like seconds, and make sure that it's going to work in a customer environment and then it's quality.
00;15;13;25 - 00;15;41;17
Unknown
It's easy to create I, POCs and toys and deploy that, on your local workstation or, you know, in Vercel or as a static web app in Azure. But getting that to the enterprise grade is, difficult journey. And, and because that that ties into things like data security, you know, could you bring down the network, all that kind of doomsday, doomsday type stuff?
00;15;41;17 - 00;16;01;21
Unknown
And so, Canisius and I should not have been drinking a bottle of wine as late as we were the other night. But we've talked about that, that I do think we're going to have to get to the place where our clients can build their own apps and put them on collide, because we're just that's the way the world's going.
00;16;01;21 - 00;16;23;06
Unknown
But we are going to have to, be the ones to make sure the system works. Yeah, yeah. It's, Yeah, I wouldn't say it's scary at all. I, I wake up motivated every day to solve the next challenge and think about, you know, what's coming next. We've got a lot of exciting things in the pipeline, and, we've got an incredible crew here.
00;16;23;06 - 00;16;55;17
Unknown
I'm really impressed. Yeah. No doubt, no doubt. Anything we should have covered that we didn't. Yeah, I think I think it's important to, you know, for us to continue thinking about where customers are going and maybe think about how they're starting to use some of the AI tools and how we can integrate that. You know, we're starting to see a lot of enterprises also dabble in cloud code.
00;16;55;19 - 00;17;17;25
Unknown
And I think we need to figure out a way to rope in some of the, skills and agents that they're creating and somehow, get that integrated with the cloud platform so we can, you know, leverage their existing work streams. That's something else. We don't want to build everything net new here. And that's something that I've been thinking about a lot.
00;17;17;28 - 00;17;42;07
Unknown
You know, we can go automate all these workflows and let's say we can save 25% of G&A, or we can allow you to double your asset base and not add any additional people, whatever the metrics. That's cool. But ultimately, at the core of what we're doing is pattern recognition. We're only getting 5 to 10% of the oil, place out.
00;17;42;10 - 00;18;07;22
Unknown
We got 90% of the oil left. And when we have these tools and we start focusing these tools on CapEx and it's you get this log signature and you get this production profile on the field, and that means refracts work or whatever the case may be, being able to empower the the business to go do that. That's where I'm getting jazzed about.
00;18;07;22 - 00;18;34;14
Unknown
And I kept thinking, that's a five year from now problem now. Yeah, it's potentially before the end of the year. We're going to have some applications starting to look at that, I'm sure. Yeah. We're actually starting to like run to some of those questions right now. And I think the first step, the foundational step is being able to, traverse, find correct answers and navigate through a dense environment.
00;18;34;16 - 00;19;03;28
Unknown
Once you're able to do that, you can start, you know, figuring out some of those harder questions that rely on different domains. And like, really, really tough environments. The thing I'm excited about is LMS and agents necessarily aren't the answer to everything nowadays. I know that's like where all the hype is focused on right now, and many people are just, you know, using dev tools to speed up their own SAS app and, and deploy it.
00;19;04;00 - 00;19;30;24
Unknown
But what I'm really interested is, what I'm interested in is using LMS to generate, machine learning code to tackle some of the problems that you're talking about. We're starting to dabble in that already. That's that's wild. Yeah. Who would who would have ever thought that? But, you know, I mean, it goes back to what we were talking about earlier, the, the show revolution was because we just suspended belief in convention.
00;19;30;24 - 00;19;54;02
Unknown
Right. Turns out all those freaking engineers were wrong about this stuff. And the ability for AI to do that on your data and in effect, just suspend convention or belief in something to see what falls out. That's going to be cool. Yeah. I'm still thoroughly convinced that we're going to figure out if you have a left handed pumper, you have more.
00;19;54;02 - 00;20;05;10
Unknown
Well, failures. We will find out someday somebody going to have to be is there causation to that? But I've always thought that, well, dude, I'm glad you're here. Yeah. Thanks so much for having me. I really appreciate it.