Agent Management at Enterprise Scale | Conor Jensen, Dataiku

Ready? Cool, ready? Yeah, let’s go. All right. Well, welcome to the Gartner data and analytics conference 2026 I’m here at the data ICU booth with Connor Jensen. He’s the global field CDO

globally, globally, in the field. Yes, the field global,

that’s right. Hey, can you tell us a little bit about data IKU, 32nd version, and then what do you do as a field?

CDO, yep. So data haiku is a development and orchestration platform for all of your data needs. So from analytics, data pipelining through machine learning agents, anybody can come into the platform and build, deploy and manage whatever data product that you want to have. We have full code and no code capabilities so that anybody is going to be able to work in that, from business users to data scientists, on top of whatever architecture you already have. So we are very flexible in working with wherever you already have your data, and then the field CDOs. So I actually was a data IQ customer. I purchased data IQ about seven years ago, about 10 years ago, came here about seven years ago. And my job, I always just like to say, is, I’ve been in my customers shoes, and my job is to help them make new mistakes and different ones, because we’ve learned a little bit along the way, but we help with strategy and operational planning for making sure that they’re able to get the most out

of the platform. Okay, well, you had a lot of big announcements yesterday. Yes, we did. Can you just what are the highlights? And then we’ll dig down, yeah.

So we announced three new products this week. Super excited about all three of them. The first one is data IQ co build, just furthering that extensibility of the ability for anybody being able to make it, bringing vibe coding into the platform. The second one was what we’re calling our reasoning systems. This is all about adding in the process layers and the context layers into the data side to be able to build like a full blown end to end solution that manages really complex sort of problems at our customers. And then the third one, and the one that I’m personally most excited about, is our agent management platform, which is what everybody’s been clamoring for this year, is how do I manage all of my agents across my infrastructure, whether wherever they’ve been deployed? How do I make sure I know that they’re performing, not just functioning, but performing. And, you know, how do I manage that effectively? So really exciting, you know, biggest product launch announcement that we’ve ever had, and a lot of really cool stuff

that’s coming. That’s great. And you got a fresh new coat of paint, new colors, beautiful branding. We have

some new branding. We’ve got a new coat of colors. Mark will be very happy to hear that. You like it. I personally like the dark green and the beige, but, yeah, it’s exciting, really. I’ve been here for seven years. It’s a long time. The company is now about 13 years old or so, so I’ve been here for quite some time. It really feels like it’s not a new company, right? Like we’re still doing the same thing we’ve always done. But, you know, with the age of Gen AI in the last couple years, you know, it’s really changed the market, right? Okay, it’s a completely different market to live in, in this world of data and AI today than it was two years ago, five years ago, 10 years ago. And, you know, I really, sort of feel like we’re really, this is our first big evolution that we’ve had as a company, where we’re really embracing what’s coming from an AI perspective and adapting to the market and doing it. So it’s a really exciting time here, you know, which is rare to come by, right?

And you know, speaking of that, you know, a whole new world. You know, we’ve all been around a while, and we like, the biggest complaint about AI is pilot purgatory. We can’t get things to production. And I see on your banner right here, you are now the platform for for AI success, your role as the field, global field, CDO, what did you observe that kind of maybe informed the positioning that something needed to be different.

So when I think about sort of the difference between building and trying and testing and prototyping versus people who are actually being successful with doing it, I’d say there’s really sort of three key things that we have seen and experienced with our customers, who are doing this right, who are doing this well, who are being successful. The first one is really streamlining what that deployment architecture looks like. So often we get caught in the I built something that works in my laptop, or it works in my dev environment or something. I have so much of that on my computer, but actually getting stuff into production when I’m done with it as a data scientist, and I hand it over to it, and we go through all of the like rigmarole to do that, like ML ops. I know we’re already now talking about ml ops and LLM ops and now agent tops and all that. But like, we haven’t solved any of that yet, right as as an industry, we just keep putting more in the backpack. Just keep putting more in the backpack. But like ML data, Ik, architecturally, is one platform from dev to production, and it’s very easy to get things out into the hands of users. And so that really has helped doing it, because, because of some of the. Got those like, you know, not just the low code no code capabilities, but the underlying architecture data, IQ, makes it so that whatever you’ve built and is functioning in that sort of prototyping environment will work in production period, right? There’s like, No, there’s no, there’s nothing there. There are hoops to jump through. We have the governance, and that works, sure, and that’s the other pieces. We have the governance in it to make sure that that second one of you know, there’s technical hurdles to getting stuff into production and having success, but organizational dynamics tend to be a much harder challenge. How do business users adapt and deal with the change management? How do you organizationally accept the risks understanding how things are changing and everything like that? So we’ve got governance built in. We have trust built in. These are the factors that are really making it be able to go out and do that. And then finally, again, the underlying architecture is such a moving target for companies that we get caught in these loops of well, the data has to be perfect before I can move anything to production or whatever, and they’re news flash. There’s no such thing as perfect data never will be, yeah, there never will be. You have to just get moving. And we enable our customers to just do that, start where you are and start moving.

Okay? And you know that getting into production, you have a nice set of logos on the wall. So, you know, I guess proof is in the pudding. The next thing, one of the analysis, was co build.

Co build, yep, what is that? So, you know, look, I hear the news just as much as everybody that nobody has to learn to code anymore. Vive, coding will solve it. Cloud code is cool. Cloud code is cool, no doubt, super cool, super cool. And vibe coding a web app, vibe coding an application, or, you know, a mobile app or something like that, really cool for those things. Vibe coding a data pipeline is so fundamentally different if I’m trying to build even just a data pipeline, or now data underpinning a dashboard or a machine learning model, or by coding an agent, how do I know that these 500,000 2000 lines of Python that it spit out actually gives me the right answer? Right? I get the summary statistics, or I get the stuff but like and I can’t see that it works. You know, if I vibe code a web app and I pull up the web app, it either looked like what I wanted it to look like it does, or I click the button and it doesn’t work, and I have to go back and do it. But if I get this thing that says, hey, this data has this many rows in it, here’s the summary statistics and here’s the model we do. How do I know how like, unless I’ve actually can see the table, unless I can see those and dig into it, and that’s where I think data I could go build. We’re bringing cloud code in. We’re bringing in those tools natively within the platform. But instead of it spitting out 2000 lines of code, it gives you the same visual workflow that data IKU has always had that made it so that you didn’t have to, you didn’t have to use to code, to use data within data. I could anyways, but now you can use plot to create things. You don’t even have to learn to do the visual stuff. You can create those things via the sort of vibe coding thing, but you have to be able to edit it. You have to be able to tweak it. You have to be able to digest it. And you can’t do that, or you can, if you know how to read 2000 lines of Python, right? I can, and I don’t want to, right? That’s the other side of it, okay?

And so it really, what I’m hearing is it adds a layer, so it’s got that visual layer. Take the code, it makes it visual. So if your preference is to be a visual coder, or low code, you can do that. If it’s we like to live in the pure Pythonista and live in that world

too well. And the thing I’d even sort of say, one of the things that I’ve always liked about data I could and CO build just starts to enhance that is that data scientists, anybody who’s working this stuff, even if you do know how to code, we don’t always do it in sort of like good coding practices. We don’t do modularity

stuff. Yeah, rightly. Data Scientist,

like, look, and I am, I am the first to bug, right? To say that, yeah, like, I can, I will make the solution work, but it’s not pretty, and it’s probably not right for production, right? But data cool will sort of pull that thing into its disparate elements and make things clear, simple, SQL or visual executable recipe versus the stuff that does need to be coded. That is, you know, the machine learning programming is whatever it needs to be. And so even for the people who can go through and do it out of 2000 lines of Python and a machine learning, you know, sort of project, there’s probably, like 40 that are like, what’s really, really important, right? The rest of it is, yeah, okay, did you pull the right data? And that’s important you need. But I can now, like, abstract that away and not have to spend my time digging through to figure out where the important cards are, all right?

Oh, that sounds that sounds amazing. So you could use that. Okay, the next one you talked about was reasoning systems. So they’re reading into decision science here,

it’s exactly right. Agents are the first step for us in being able to make decisions autonomously, having getting out of it being I’m a human. I pull up my chatbot, I pull up my agent window, and I start asking it questions. How do we start putting that stuff in the background so that the agent can make. Decisions. But that requires a lot more than just data, right? It requires the agent reasoning, and it requires access to the right data and the right systems to do that, but it also needs the process flows. And the understanding of this is not, you know, I have five different, I mean, simple system, five different data sources that I need to be able to sort of dig to to understand my, how my supply chain, you know, is working. You know, today an analyst is doing that, and it’s an analyst who has to make that decision, because they need to pull data out of this system, look at it. Maybe they have access to all five systems, or more often, they have access to some. They talk to another team. We do all this. And so with the reasoning systems, we diagram and we implement all the process flows into the system alongside the data flows or and I mean, maybe not even actual data flows, but just the data sources that are necessary to support that process. And then give an agent the ability to sort of go through that whole process and pull together the problem at the end state. You know, that’s three, four different layers up that you have to be able to understand this data flows through to here, flows through to here. And we have RPA. We have, you know, these tools that help us automate processes. And some things are RPA able, and some of them are not. Some of them require it’s not deterministic all the way along the way, and that’s where we’re sort of pulling that logic that comes from an RPA style system with access to data and models and tools under the hood and an agent that then can walk all that through and go, No, I shouldn’t do this next step because I don’t have the right results. Or, you know, I should just, I should stop it, or I should throw self correcting. And these are, we’re doing these. They’re very targeted, right? So we have industry solutions teams that work with a handful of industries, and we are building these sort of, like, one by one for specific use cases and specific industries to solve some of these, like, really gnarly use cases. I think

that’s the key, is, you know, you got to be focused. You can’t just, like, AI’s not gonna, it’s not like the magic bullet. It’s not gonna solve all the problems with your targeted approach. I guess, smart way to do things. Yeah. Okay, and the last thing you mentioned, system to monitor agents. Yep, is that what it does? So probably does a whole lot more.

Yeah, definitely not just monitoring. So the this is, and this is like the number one conversation I’ve had in the last 12 months with our, you know, the leaders and our customers, but also CIOs, CDOs, you know, in my peers and other companies is so I have data IKU, and I can build agents and data IKU, and they are, of course, kick ass agents that you can build in data IKU. But I also have Databricks, which has kick ass agents, sure, and I have Salesforce for being in the agents. And everybody’s got an agent ecosystem. Everything is an agent ecosystem, right? Every company that used to be a data company is now an AI company is now an agent company. But how do I know that I should build an agent versus use the agent that’s inherent and native to the platform that I have? How do I know as I’m standing up these agents from all these different systems, I mean, first question for most CEOs is, how many agents do I have? That’s actually a really hard question to

answer today, right? So we take in the problem, like, I don’t know what my data is, and apply it,

and we are just scaling it right, right? So that’s the first step. Is just create that observability to be able to say, here are how many agents I have in production across all of my systems. That’s the first step. But now, are they working? Are they working mechanically, the monitoring, right? Sure you know, what is my uptime? Am I getting satisfactory? Where are those things? That’s important, but you can do that with an API bus. You don’t need something fancy to do that. But the most dangerous thing, or a far more dangerous thing than an agent that breaks, is an agent that’s still functioning and giving the wrong answers. Okay, so you have to then add in that layer of the performance management of the algorithmic, agentic performance management of understanding. Is this agent working? Yes or No. Is it providing the right results, the right answer, the right whatever it is. How do I understand whether or not these agents are giving me the business results that I need? So it’s adding in the semantic layer, it’s adding in the context. And it’s the ability to do this across all of my platforms that I can say, across my eight, 910, 20 platforms that are out there. You know, I have 10,000 agents in production. Are they all functioning? How are they doing? Where should I be sending something back to a team? Where do I need to take stuff offline? Because we’re, you know, it’s, it’s gone haywire. That’s the idea behind it. The cool thing about data IKU and the because, hey, could any company go out there today and sort of build this dashboard that pulls these things all together? Okay? Maybe not any company. Most companies have the ability to do that fine, but there is no shared look. MCP is great. Hoa is great. And I’m certainly very hopeful that we do find ourselves in a standard, that there is an HTTP, that there. Is, you know, like that typical protocol that we all align on, but we’re not there yet, and we’re a couple of years from that. So now either you wait until that’s here and then you start monitoring the stuff, or you build 812, 20 integrations into a platform to pull out whatever they’re surfacing from an agentic perspective, some of which is levering, some MCP, some a to a, some home baked whatever, like we’re doing that work. We’ve got those integrations already because of the platform that we are. So now we’re just extending that to the agentic stuff, pulling all these things together and making it so that, you know, a company can look across and see what’s where is it working? How do I make where do i Where is the agent from? One of my systems just not that great, and I should go build my own versus, hey, these things that are coming out of Salesforce are great. They’re doing the job. And I, I can go focus elsewhere. This is, this is so exciting for me, because this is, you know? So we talked about the all the projects that don’t make it to production, and it is a problem. We’ve been talking about the data side for world for years and years. Well, with the ease of building with agents and the ability to vibe code and all that sort of stuff, we’re not solving that problem. We’re just expanding the backlog, right? And the ability back to that sort of like that fear and that trust of why a CIO isn’t letting these you know, whatever models we’ve talked about for years, go to production, is because of the difficulty of seeing them, the difficulty of monitoring, the difficulty of governance, once you have a platform that allows you to see everything that is moving to production, tie it into your governance to make sure that the right steps were taken along the way. Have the monitoring on there. That’s what allows us to sort of relax the guardrail. I shouldn’t say relax the guardrails. Open the gates with the guardrails to let stuff sort of finally flow through.

Here’s a sandbox you can play in. Go play. Yeah, I know what I’m hearing,

go play, but then it’s really, click, click, great. Now it’s gone off. I’ve tested it through its processes, and now it’s out in the wild, and I know that it’s working. Nobody’s gonna suddenly come back to me and go like, Hey, how did we lose all of this money? What happened? Right, right?

It’s there. You know, it sounds extremely powerful. So maybe just one final question, yeah, from what I understand, it’s sort of, it’s part of the data ICU platform, or is it different? It’s a different

UI, it’s a brand made a deliberate design choice. Yes, great question. So obviously, it leverages some of our technology, but it is not dependent on being a data IKU customer. It is, is coming, to be fair, right? So, you know, we have early customers who are in early adoption with it right now. So that is existing customers only. The plan is it for it to go GA in September. That’s our target release date right now, and that will absolutely be available to any company to be able to stand this up. It does not require you to use data IQ, so you can manage all of your agents across all of your other platforms, even if you’re foolishly not building them on data Ico while

you’re out. Well, I love that. Connor Jensen, global CDO of data Ico, thanks. Thanks for the discussion. Really appreciate. David, Bye.

Cheers. Thank you, sir, my pleasure. Thank you. Thank you. Yeah, for more of these, I’m game. You.