981: How Data Engineers Are “10x’ing” Themselves With Agents, feat. Matt Glickman

26:47And, of course, they're not going to run every single report because there's humans, right? So they'll pick a sampling and they'll try to see. And then worse, they'll make the business users be the testers, right? Because they know what really matters, not all 30 ,000. With an AI, you'll run every single one of those 30 ,000. And you'll know every single thing that's wrong. And it'll be very, very thoughtful and very great documentation on how to cover it.

27:15Jon Krohn:Yeah, people worry about error rates with agents or with LLMs in general. it seems to me like if you're willing to spend on the tokens to be double checking triple checking things you pretty easily like you talk there about checking over every field every single item that goes into a report every line of code if you're willing to spend on the tokens to double check triple check everything you're going to very quickly surpass especially since it's february moment that we've been talking about you're going to quickly surpass human accuracy levels on i think basically on a very, very wide range of tasks, including data engineering.

27:51Yeah, no, definitely. But I think we also, we are giving humans way too much credit. Humans make mistakes all the time. And it's just accepted as like, well, we're only human. Right. And yet we raise the bar saying, I don't know, but this other thing that we've created that somehow is like us. It's a well-documented effect. Right? We just expect it to be better. So number one. Number two is, it is something to be concerned about. Right? And I think just saying that I'm going to single shot and YOLO and throw some tokens at this and hope gets the right answer is not the answer either. And part of what we've invested and what we've been building over the last now two years is a harness that keeps these models on track to not only verify that they get to the right place when they say they are, but actually make them prove it.

28:47Part of this next wave is where these models understand how confident they are in their responses. So using that and holding them to task of saying, okay, you said you're confident, provide me the actual artifacts that prove that you did these tests, that prove that you ran every one of these 30 ,000 because for reasons that we'll somehow, you know, maybe or maybe won't understand, the human laziness factor has made it into these models where they will do the minimum necessary to get the answer, right? Because you don't want to waste tokens. You don't want to be just going on and on and on if you don't need to.

29:36And if it's like, well, I don't, you know, I don't really need to run every three, you know of those 30 000 reports so i'm gonna just you know say it looks good right versus saying like i want proof i want you to actually give me the results give me the outputs and then we'll have another ai that will review it and say like no no no sure there's only there's only 10 000 there um so that kind of checking has been a big part that combined with leveraging this effectively this internal knowledge graph, this internal context graph, and having that be the kind of the two pillars of guiding this intelligent being that we've all created is what's making this work in enterprise.

30:20Jon Krohn:It sounds like a known brainer to be taking on these kinds of data engineering agents to help myself and the audience better understand how this works in practice. Are you able to walk us through one or two use cases, maybe anonymized use cases with clients of yours in terms of how you implemented your solution and what the impact has been on that business? I mean, part of, I guess, the third pillar that I'll kind of mention and then I'll take you through the example is we're agnostic to the actual data engineering tools and platforms that people are using. These platforms will come and go. The frameworks will come and go.

31:05What we're replacing or really augmenting is the people. So now everyone out, to be clear out there, anyone who's panicking, no one's losing their job. I had that question coming up. No one's losing their job. What's going to happen, though, is that people are not going to be hired as much. So the next wave of people who would have been the junior data engineer or the junior analyst or the junior operator or just honestly any junior role in enterprises if you want to get kind of a little more dystopian about it. But the junior lever hiring is the problem that AI is basically going to wipe out.

31:41And it's already starting to happen. Hiring productions are just not going to happen. People have the skills already there. There's no reason to replace them, particularly in the data space because they were limiting the fact they just didn't have enough of them. Now you just can get 10x more power out of them and they can move on to things that they wanted to do and not the tasks that they were basically bogged down to do.

32:02Jon Krohn:Yeah, maybe I'm being too optimistic here, but don't you think there are some scenarios in some organizations where they actually will want to do more hiring because, and even of junior people, because they're being so much more productive with the individuals that they have. They're getting so much, you know, like you said, a kind of 10xing the capability there. You were talking about how previously data engineer was this extremely stressful role. They never felt like they had enough time for anything. and I suspect that even if you're 10xing your capability, there's still a lot to be keeping an eye on and making sure of.

32:37Jon Krohn:And so as that ecosystem grows, you can imagine some organizations also being like, you know, it'd be good to have a few more eyes on this. Yeah, no, I think that the few more is the key part of this because there's, my co-founder Justin published a piece that if you haven't seen on our, it's on our LinkedIn, on Genesis LinkedIn. There's going to be certain people who can actually be these kind of conductors of these AIs at scale and actually use them better than others. He calls it spinning plates. He now has like six different kind of agents working in parallel on different things that we're building.

33:21I've only gotten to like two and a half. It's like, so there's going to be certain people who can come in and be these force multipliers and say like, maybe you're working with one agent is doing something with you. I can actually command 10 of them and I'll have the context switching in my head to figure out how to put them all to work. So there'll be some of that hiring. But that's, in my opinion, that's going to be kind of the few special hires because the number of people who can do that just a limited kind of audience. I think it's more about just the impact it's going to have. Everyone who's doing these kind of operational roles are going to be the ones that now just get a force multiplier applied to them.

34:14And I just see it hard-pressed to want to bring on more humans when you can just one of 10 agents, 20 agents do that task instead.

34:25Jon Krohn:It wouldn't be surprising to me, and I'm open to your critical feedback on this, but it seems to me like some junior hires might be more likely to be that kind of 10x agent orchestrator as opposed to 2.5x because they could be growing up in this ecosystem where it's like vibe coding first. I hope so. The thing I worry about, and this is maybe a little off topic, but I think it's relevant, is that the education system has not yet figured out how to teach AI. Yeah, a lot of places haven't. Some places are. And I think if that happens, maybe this can change dramatically. But right now, most schools do not, in high school and college, discourage people from using AI as part of their process.

35:22So it's like a double whammy. Basically you're going to have companies that are just, if anything, like I said, if anything, they're looking at people who are masters of how to use this tech. Any people coming out of school that have learned it despite their schools kind of not encouraging it right um so yeah it's a bit of a double whammy and it's going to happen i mean i i'm you know i'm actually trying to give back and work with you know the schools that you know my kids have gone to and just to try to express the urgency um and i think there's appetite but it's tough i mean it's how do you incorporate this how do you how do you assign papers like what does that mean like the idea that you're gonna be able to tell you're not that that's a farce No human would be able to tell, and it's just going to get worse and worse.

36:08Jon Krohn:Yeah, yeah. To bring listeners into kind of an arc of conversations that we've been having about this AI education problem, the human education with AI problem, is that a couple of episodes ago, in episode 977, we had an NYU professor, Kyungyung Cho, on the show. And he has his, he's teaching undergrad machine learning intro course. and for the first time it's vibe coding first. And he said that it's surprising how many computer science students at NYU have limited to no experience using these tools. And you think those would be the first adopters. But on the note of even younger people, people kind of K to 12, and what are we gonna do with education for them?

36:55Jon Krohn:Really interesting episode, 975 a few weeks ago with Zach Cass was exceptional on that. And then we have, I haven't recorded it yet, But I'm expecting that the very next Tuesday episode is going to be with a K-12 educator who specializes in trying to get some kind of early adoption of these technologies. That's awesome. So hopefully we'll have some answers for you parents or soon-to-be parents out there. I hope so. Because the dislocation is going to be harsh. Yeah, I hope there is. But back to the question of like, so that socioeconomic problem aside, basically how we engage with customers is basically they onboard Genesis as if they were onboarding new employees on their team, right?

37:43So they connect the system to their platforms, give them credentials to read from their document repository, read from their databases, read from their code repositories. And then they basically define effectively a project or a mission that they want the agents to go on. And the example would be I want to basically build. I'm an asset manager and I want to basically understand all of my assets by client type, by asset type. I have some raw feeds of things I'm getting from my custodians or my banks. and I want to understand everything and create everything I need to actually have an interactive dashboard that I can slice and dice with all my different attributes, which would have been a mess.

38:33Literally, my team built that kind of thing. It's a massive undertaking of just gathering all that data, normalizing it, combining it, linking it, putting all the business logic that typically was hidden away in other applications now into these kind of data flows. producing it as some kind of output. But basically, we've had it where you can literally say like a hand-drawn diagram. This is generally what I want, which actually happens typically. If someone who understands the business says, I want to have these kind of charts and this like, here, just go and build this and come back to me when you have something.

39:10And they come back, whatever, weeks later, months later,

39:13Jon Krohn:and it's half right, half wrong. Basically start there, and the system basically starts going. The agents will go and introspect, understand what's already there, understand any kind of, like I said, documentation, code, or anything as part of that context graph, and starts building. Building and testing and validating and iterating. Taking the best of the coding agent models, but also all this kind of context, and using these guardrails that we call blueprints. So we have a set of these, as I was describing, kind of guardrails that we say, like, if you're going to do, if you're going to extract data, this is the kind of runbook you want to use.

39:59You want to extract it, validate it, confirm it, and create a monitor in place to make sure it's always going to be fresh. If you're going to be translating data into some semantic model, you know, called a source to target mapping, you're going to basically go field by field and make sure everything ties out and all these kind of things. and each step along the way. And agents are basically doing that on its own. And the key difference, coming back to AI first, is that instead of being a co-pilot, you have to say, okay, now do this. Oh, wow, that was pretty impressive. Now do that. Oh, you missed it.

40:31You should go back and do this. We've reversed it. Instead, have the AI is going, working on a task, and when they're not confident or they get stuck, they then come back to the human. Right. And the important thing is when they come back and they say, no, no, no, this is when I say revenue, this is what I mean by revenue. That thing gets memorialized for next time. At the end of the day, but you can now go, we have these now projects that go on for hours where it's going on and doing things on its own. And, you know, stopping minimally. And particularly as they do this more and more, they can really go all the way through.

41:09but it's a combination of that context this kind of these harnesses, these kind of blueprints that they can keep on track and it's the fact that they can then learn as they go and now become the center of knowledge but at the end of the day that's cool and all but ultimately they solve the problem these are not assistants these are great data engineers that are just you now can scale up on demand. Right. Nice.

41:42Jon Krohn:And so your clients have been concentrated primarily in like finance, healthcare, or is there some kind of special? healthcare is, is dominating. We've been able to, we've been able to, to attract a surprisingly normally hard to hard to onboard set of enterprise customers, mainly because of the way we've chosen to deploy. I have a new appreciation for what Snowflake accomplished back in the day of convincing these enterprises to let their data leave into a SaaS that now had to become a trusted entity. That just doesn't happen much, definitely not anymore. So the way we deploy, which is definitely the harder way to do it, we deploy Genesis into their environment.

42:35In traditional world, you would basically give them software and they have to install it and manage it. But the magic of AI, we're basically giving them something to install with an AI engineer inside. Who you basically say, okay, run this command and now the system is going to effectively manage itself. And even give back feedback on things it's learning on site. But then it's all in the control of the company. it's now their asset that's accumulating. We don't get any knowledge that's accumulating. We don't get any kind of, even telemetry they don't have to give us. And now they're willing to expose it to everything because it's something that's secure in their barrier.

43:18So that's like the third kind of, I guess, third kind of leg of the stool of being able to have something where it's purely trusted because it's running inside. but we can do that now because it's an AI-powered engineer inside that is managing the system as well as accomplishing the task at hand.

43:38Jon Krohn:One of the key elements that I think allows that to work for you, allows it to work for Genesys as well as your clients, is something that you mentioned already earlier, but I want to highlight how important this is. You have a blog post that we'll link to in the show notes called How Genesys Automates Data Pipeline Development in Hours. And in that, you talk about how the agents escalate when confidence is low rather than force an answer. Tell us more about that and how important that is. This was a big thing that we realized that we had crossed. One of these kind of massive steps forward was up until, and this really happened when the reasoning models landed early last year, which everyone is excited about and basically show that you could scale up inference time thinking.

44:28But what came out of that was the ability for the models to be guided, to be much more self-reflective on an answer. So you give an answer. And up until then, you could coax them. You could threaten them with violence, which I still think is a terrible idea. That'll come back to haunt us when Terminators come. But there's nothing you can do to basically try to really get them to say, like, how confident are you?

44:54Jon Krohn:Yeah, there was an interesting study that showed that saying really aggressive things actually gets you like 10 % more accurate responses. I know, which is, to me, it's just not worth the 10 % to be on that list. But we saw that the real big win that was not often talked about was this confidence indicator, where you could basically like, how confident are you that you did everything I told you? And it would be very good at telling you, I'm 90 % confident. And this is why, and this is the one thing that I'm not confident about that I need clarification on. So that was a big kind of moment for us.

45:28And we basically now harness that. So every step of the way, when we're going on these complex projects, complex missions, we're constantly asking, okay, you did that, right? And how confident are you that you did it correctly? And again, if you are, great, show me the artifact. If you're not, then go back and try again. but when you're not, escalate what are the missing pieces, right? And with those, that becomes like, well, you know, I had to guess what this formula was because I just couldn't find it, right? Which if you don't ask for, it's just going to be something that goes under the radar.

46:07Again, coming back to like, you know, the human, we hold the humans to a higher bar, humans do this all the time, right? You basically just make these jumps of logic. You're like, I think that looks right. I'm going to go with that. And now someone asks you, how did you come up with that? Well, I actually made that up. It sounded good. They do the same thing. But if you call them on it, you actually get it to be much more productive. And then they basically can ask you the intelligent questions. And the most important thing, and this is nothing more frustrating than when you're asked a question and you basically give the answer, you really want that to be applied next time.

46:45like nothing's more frustrating without you dealing with a human or an AI where it's like, you asked me a question, I gave you the answer, you said that was a great observation. Don't come back to me tomorrow and say like, hey, what do you think about this? It's like, I told you that yesterday. If you do that right and you capture those moments and those nuggets and you do it in a secure way, like now, you know, now it's, there's no limit on what you can take on and it compounds, right? Because now once you understand all this logic and how these businesses operate, you can then move up the stack because now you understand all the semantic, all the flows, all the kind of how we got here.

47:25And now we're finding that our customers are pulling us to go further up stack because, well, that's great now. Can you help me actually present that to the board? Sure. And now we have our systems actually able to produce a well-thought-out presentation because it's grounded in the facts of how that actually operates.

47:45Jon Krohn:Makes a lot of sense. And to dig into that just a little bit more on this correctness point, you've previously in an interview stated how with consumer-facing products, novelty is often one of the most important characteristics of AI systems. You're talking there about some kind of inventiveness that these models tend to have filling in the blanks. That novelty piece is key for consumer-facing products, but for enterprise products, it's correctness. Yes. Correctness is everything grounded in truth and the ability to navigate these complex organizations and extract out like what is correct. The best run organizations do not have the rule book that's like this is all the things that are correct.

48:31They just don't. It's in people's heads. Some of it's in coding systems. still to this day, why do people stay at these regulated industries for so, such long careers? Because of all this knowledge that gets stored up here, it's cheaper to keep that person and keep on paying them than trying to download it out of them and put it somewhere else. Again, I was, you know, a Goldman for 25 years and that was normal, right? Still to this day, I mean, people stay there, this is going to change, right? Because all that knowledge is going to become, what was a liability for companies is now going to become an asset where now if I can have, imagine if I have this system that has all the knowledge about all of how a major banker or healthcare company, thousands and thousands of employees now have put all this knowledge implicitly in a knowledge base that the company owns.

49:27I mean, you're going to see companies, you're going to see M &A, that a variable of the M &A is going to be like, well, do they have a consolidated, you know, knowledge-based context graph of how the firm operates? I'm going to value that more versus like trying to, because people are going to leave, right? What happens every time we have an M &A and it's like risk, you know, people are going to leave, who's going to keep it, who's going to pay.

49:49Jon Krohn:But it's like, yeah, it's, we got it all. It's like the classic of how you get a much better multiple on a SaaS product business relative to a consulting firm. Exactly. Because the consulting firm has so much. Yeah. human knowledge that leaves with the people. Speaking of these knowledge bases, you call them living context graphs. And so these are systems designed so that this institutional knowledge compounds over time and is never lost. And it looks like this is born from your experience watching critical organization knowledge disappear across data teams at Goldman Sachs and at Snowflake. So can you tell us more?

50:24Jon Krohn:Obviously, you can't get into too much about your secret sauce, but we have a technical audience. I'm sure they'd love to hear a bit more about how these living context graphs work. No, and it's interesting that we fell into it. We didn't sort of say, you know, we're going to build this agent platform. We need to build a context graph. It was basically we realized that the missing piece that we were constantly feeding all these agents to kind of get going was a bunch of context. And that was the big human element was like, okay human go and find all the documents find all the you know repos point me all the relevant databases and then it would go and be super successful and we all say that's not going to scale because no one wants to want to do that gathering because now it's a big you've now took take off some of the work but you've made you're making the humans do more or other kind of hard work so what we basically do as part of genesis onboarding is you connect the system to all your databases, all your repos, all your SaaS tools, or as many as you want to, or on-prem, in-cloud, wherever, and then they go about and start crawling.

51:35So think about a traditional kind of web crawler, but in the context of wanting to understand all the data relationships that exist amongst an organization. A spreadsheet here, a database there, an API call here, like all the things and then effectively layer on top of each other. So now you can see that this code is referencing these tables and these APIs. And now you can effectively build up a graph. And you can see it. We have a demo on our website. You can see it how it literally becomes almost looking like a social graph, but of the data relationships amongst the firm. And it gets super complex, super fast.

52:12So the goal is not to have any human ever get their head around it. But an AI, again, loves this stuff. And the crazy thing about it was that all we did was we did this crawl. We got all this kind of built up of this graph and all this kind of metadata on it. And we just gave our agents tools to navigate this graph without even explaining why. We've since now kind of guarded it. I mean, guard it a little bit. But it instantly said, like, this is great. I can now understand. And it was like we'd given it the secret formula. And it basically just started crawling. and say, for example, a user wants to add a new column to a table on a report.

52:57Our agents basically went and Eve is our master agent, Genesis, Eve, Adam. Seemed clever at the time. Basically, Eve will go and say, okay, well, where am I going to get this data from? Let me see where there's similar data around. It goes and figures, how did I produced this report so it understands that kind of path it's like can I follow that same path back to its source and it wasn't was there missing field yes or no maybe or is there some other field or some other reports similar that I can find and basically it's you know search on this you know multi-dimensional graph space and try to find a similar semantically similar search on the space and find something similar that I can connect to and then pull that data it through and build the pipeline.

53:48So it just fell out of that. And this, these, you know, the AI was, you know, it was almost like, you know, it was meant to be, but this was going to become, and now it's been self-fulfilling where now everyone's figured this out and, you know, it's getting better and better and don't know how to navigate this. So, yeah. And that, that basically was a huge unlock for us. And it makes our systems come online that much faster, that much more productively and removes the overhead for humans to kind of train the system or effectively similar to how like you know Google went and crawled the internet they didn't you know have to ask people to go to Yahoo and put in you know links and you know and maintain that they just did it passively and uncovered all that relationships on the fly doing the same thing but we're not doing it to produce a graph we're doing it to do better data data project and data engineering.

54:46And it's because of that reasoning, people are willing to kind of connect more and more systems because they see the value.

54:52Jon Krohn:Yeah. You create the graph to enable a better AI system to be able to crawl that graph and have knowledge more quickly, better, more concretely represented ideas. Cool. All right. So if people are listening and they're thinking, I'd love to have this personally. It's obvious. Or in my organization. A key part of the adoption problem here is that it seems like a lot of people are still thinking about the pre-February, 2026 mindset of is this an AI use case instead of why shouldn't this be done with AI? And so you outlined a four-phase adoption model that begins with assessment and ends with scaled autonomy.

55:38Jon Krohn:Do you know what I'm talking about? Yes. Can you tell us about that? Yes, yes. No, I think you have to be thinking about how you're going to, how you're going to get there and what are the steps along the way. Like you're not going to just turn a system on and it's going to be fully autonomous, nor would you want it to be because you won't. It's actually more for us humans to come along on the journey than the AIs who may be ready for it before we are. So it's understanding the problem, being able to kind of understand what would a human do to solve that problem and then be able to get there in a way where when the AI is successful, we understand how it's achieved it and then we're willing to let go.

56:30And I think like that kind of, but it's really more about the humans being comfortable about it. But you have to constantly be trying to push the limits because the space is evolving faster than we all thought. But also don't be fooled by these amazing coding agents

56:55who inspire us to do things, but without the right guardrails and without the right kind of context, contextual understanding, they can cause more damage than good. But with the right guardrails and the right context and the right human oversight, it's a wonderful time to be in this space.

57:23Jon Krohn:How do you convince your enterprise clients, these big organizations with lots of liability risk and lots of people internally who are probably skeptical of what AI agents are capable of doing today, how do you convince them to make the leap from chatting continuously with a conversational agent to delegating to a team of agents, where say Eve, as you've described it, is your head agent kind of orchestrating. how do you convince your clients that that kind of delegation and trust is the right time? By showing not telling. Like in the end of the day the pain is so high in these places the demand is so high and been so underserved for so long that the answer is always like if this works it's a no-brainer and doing it in a way that it's doing things, it's all audited, it's all documented, it's not running rampant.

58:32It's going to follow your normal processes. It's going to test things. It's going to run in development. It's going to provide a code review and a PR to submit to your CICD pipeline. It's no worse and I argue better than you hire a new employee because you're going to have processes that prevent that new employee from going and crashing production, right? It's basically as much as risk as that. So the only thing they have to lose is that it might not work. But they all have such long backlogs. They have so much pressure to do more with less. And almost everyone sees this opportunity to basically now get out from behind the curtain and say, like, let me show you what this can do.

59:21I can actually focus on a business-impacting goal instead of working on this machinery. So this is going to pent-up demand, but the risk is it's as they already have processes in place that prevent rogue developers from doing wrong. Eve just signs on as just another developer on the team. So the only opposite we've had is people don't believe that it works because they're like, either they've had their own experiences or they've tried one of these coding agents or they've, you know, worse tried to build their own agents, which, you know, I think is also, you know, a fallacy, like focus on problems, focus on outcomes, right?

1:00:02You don't, traditionally, you didn't hire, you know, a consultant and like figure out, you know, can they give them tests on how they can, you basically like, you know, or you said like, I need you to do this migration, right? And they said, okay, well, this is what's going to cost and this is the people and this is how we're going to do it. And you would compare different options. You didn't care how they were going to do it. You didn't care which people they're going to use. And you were selling outcomes, right? Similarly, like that's what businesses want. And enterprises are getting even more now kind of critical where like if it's not core to their business, back to my earlier point, right?

1:00:44they know doing it for revenue will win so why not pick a winner instead of trying to kind of keep up with something that is just going to be you know accelerating out of their reach right yeah gotta stay on top of this fast moving thing for sure you described two moments in your career first with snowflake and later with genesis when you realized you could either sit around and hope

1:01:10Jon Krohn:or you could actively help shape what came next. How do you know when a technology shift, like the one that I think we are both convinced we're in and hopefully a lot of listeners as well, how do you know when that technology shift is real enough that you should stop analyzing it and start building for it? Yeah. No, I think I truly believe we're all here on this planet or wherever planet we end up going to for a purpose. And if you're kind of self-aware enough, you kind of know what you're here to solve. To unleash the machines. To unleash the machines or just more even a meta problem. Like for me, my entire career has been about unlocking this limitation of there's not enough people who can actually understand technology, understand a business problem, and kind of connect it to.

1:02:02And if I look at a higher level, I've been trying to solve the problem forever. with platforms, with going to Snowflake and trying to provide that as a capability other people could use. And now with Genesis, basically kind of unleashing that with AI, I think just taking a step back in your day-to-day to understand what is the world trying to tell me. And you're going to have these moments, like my meeting with Benoit when I came to Goldman, or when I had my moment where, you know, the early GPT-4 basically explained to me before anyone else was talking about agents that it could call functions on my behalf or I can call functions on its behalf, you know, as a way to basically what agents became.

1:02:54Like being aware of these moments and saying like, well, go, or Mike, again, the customers I was talking to were constantly saying that they were trying to basically democratize, you know, their data teams. be aware like what is the world trying to tell me here and and do i have an earlier view on where this is going to go than the rest of the world right if i if i'm behind that's not the time to jump in and start doing this right if you're ahead and you know this is in your wheelhouse because of something that you've seen that maybe the rest of the world hasn't seen. That's likely the time to jump in.

1:03:34The challenge is that these moments are now kind of, they're not, the openings are smaller now because of the pace that we're in. Like we are in the event horizon. Anyone who doubts it is clearly not touching the space, right? It is exactly like, you know, Kurzweil predicted. We're in the exponential. It's accelerating. You can no longer track you know you know the the the improvements there's no plateau happening anytime soon scaling is continuing to scale um so the only challenge in this kind of approach is that you have to be more aware you used to have like openings and you'd see them and they would present themselves you have time to think about it now it's going to be things where it's going to happen and it's going to be like you know instead of being like a month opportunity it's going to be a much shorter time frame.

1:04:26But I think if given what you can now do and build and try in the shorter periods of time, I mean, you can in a weekend, right? The guy who built OpenClaw did it in a weekend, right? And that changed the entire game on personal agents, right? So I think trust your instincts but always ask yourself, why am I seeing this before other people? Because if you can answer that question, then you should jump in.

1:04:53Jon Krohn:Really cool. Great guidance there. Thanks so much, Matt Glickman, for that guidance at the end. And some excitement, anxiety, how can we get on top of this so quickly? Yeah, it's a really... Our brain registers the same neural response for excitement and anxiety, and then it's up to your cortex to interpret that sensation. That's pretty cool. And hopefully most of us are taking a step back and taking the opportunity that we're in this event horizon and there's lots of exciting things we can be doing. Let's take a step back, use this incredible tooling to build an open claw type thing in a weekend because you can do it now.

1:05:40Jon Krohn:And make a huge impact. So yeah, whether it's with adopting Genesis as an AI data engineer within your organization or building something yourself, very exciting times. Thank you so much for sharing so much knowledge that you've accumulated over these decades. Really appreciate it, Matt. Before I let you go, I always ask my guests for a book recommendation. No question. Hitchhiker's Guide to the Galaxy by Douglas Adams. If you haven't read it, you should read it immediately. If you've read it, you should read it again. it is uncanny how the entire AI explosion we're going through was predicted with with such art with such precision in a comical way and and it's basically they try to build a supercomputer that's the super AI and of course I won't spoil the punchline but yeah it's it exactly we're living through what Douglas Adams predicted.

1:06:38Jon Krohn:Yeah, I don't want to, I think I can say, I think I can make this point without giving anything away, but something that's very different about the supercomputer that they're building there is that it takes a very long time to compute, like to do the kind of the big inference, whereas it seems like something very different about what we're going through now. Don't assume we're at the end of the book yet, right? because i mean think about what we're trying to do right if you even listen to what what they're what xai is trying to do with croc like i mean it's i mean there it's like art imitating reality because you know they're actually thinking about that book a lot um elon is when he's when he's funding xai but like you know what we're doing now is it could be the the early stages of the of the big you know computer that is built um you know we have not gotten to the point where any of these systems are actually discovering new things yet.

1:07:36So I think we're really early in that buildup. For those of you who didn't catch it, I think it was last weekend, Knuth, who's one of the famous computer scientists who basically kind of disappeared after basically defining how we should do computer science. He's like 90 years old, and he just published a paper that he co-wrote with an AI about a mathematical problem that he had not seen a solution for yet. So that's like maybe one of the first examples that we're approaching this new place that the book plans about.

1:08:14Jon Krohn:I think open AI researchers have been talking about 2026 or 2027 making kind of new physics discoveries they anticipate. It's going to happen. And what's interesting, I'll leave the audiences with some interesting, most interesting experiment that I've heard that is going to be done, which is basically to roll back all the training data to what was available to Einstein at the time, but not anything else that was published in science or anything else. And then see if that kind of time traveled model can produce the theory of relativity. that's going to be the ultimate test. That's a fun idea. Yeah, it's super interesting.

1:08:57And I mean, it's, I don't know, it seemed pretty hard to do, to isolate, you know, not like no written word, no newspapers or nothing. Yeah. But if you can do it, I think that's going to prove that we're over there. For sure, yeah. Leakage could be a key problem. Leakage is a problem. Yeah, exactly. Interesting times though. So awesome.

1:09:12Jon Krohn:Interesting times for sure. Matt, for people who want more of your insights or more information on Genesis after this episode, how do they follow you? Yeah. So website, genesiscomputing.ai, also.com. which was an interesting purchase. And I'm on Twitter, Matthew Glickman, and on LinkedIn, Genesis Computing or Matt Glickman. There is a doppelganger out there where I finally actually crossed paths with, I am not the West Coast Matt Glickman. I am the East Coast Matt Glickman. I think if I remember correctly on LinkedIn, you're Matthew J. Yes, I am Matthew J. Yes, yes, yes. Just to have some, you know, separation.

1:09:48Jon Krohn:Yeah, there were a couple times before we booked you for the episode where Natalie on my team showed me that other, the Duffelganger one. Because yeah, because he's in tech as well, right? So you're kind of like, is that him? No, that's not him. Yep, I'm the other guy. Nice, all right, Matt. Thank you so much for coming to record with me in person. This was a really interesting episode. Thanks for having me. Really exciting times. Awesome, thanks a lot.

1:10:12Jon Krohn:Lots of food for thought. In today's episode with Matt Glickman in it, he covered how February, 2026, marked the moment the latest Frontier models crossed a threshold where they could handle complex, multi-step data engineering workflows that previously required human expertise, and this big change means there's no going back. He talked about also how finance and healthcare were late to adopt the cloud, but are among the earliest and most aggressive adopters of AI. How Genesis Computing deploys its agentic platform directly inside a client's environment, more like onboarding a new employee than adopting a SaaS product, so that all accumulated knowledge remains the company's asset.

1:10:49Jon Krohn:And he talked about how, rather than acting as a co-pilot that waits for human instruction step-by-step, Genesis inverts the model. Agents work autonomously on complex data engineering tasks, only escalating to humans when their confidence is low, memorializing every answer so they never ask the same question twice. As always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show, the URLs for Matt's social media profiles, as well as my own, at superdatascience.com slash 981. All right, that's it. Thanks to everyone on the Super Data Science Podcast team, our podcast manager, Sonia Breivich, media editor, Mario Pombo, partnerships manager, Natalie Zajski, researcher, Serge Massis, writer, Dr.

1:11:31Jon Krohn:Zarikar Shea, and our founder, Kirill Aromenko. Thanks to all of them for producing another stellar episode for us today. For enabling that super team to create this free podcast for you, we're deeply grateful to our sponsors. You can support the show by checking out our sponsors' links, or if you'd ever like to sponsor an episode yourself, you can get the details on how by making your way to johnkrone.com slash podcast. Otherwise, please help us out by sharing this episode with people who would love to hear it. Review it on your favorite podcasting app or on YouTube. If you write a written review on Apple Podcasts, I will read that on air in an upcoming episode.

1:12:09Jon Krohn:Obviously subscribe if you're not already a subscriber, but most importantly, I just hope you'll keep on tuning in. I'm so grateful to have you listening and I hope I can continue to make episodes you love for years and years to come. Till next time, keep on rocking it out there. And I'm looking forward to enjoying another round of the Super Data Science Podcast with you very soon.