Happy.
Hey. Hey.
Hey. Hey.
for clear.
Hey, hey, hey.
Ooh.
Hey. Hey.
Hey, hey, hey.
Hey, hey, hey.
I'm...
What up?
Every success starts as a decision. Every decision starts as a signal. How fast, how far, how sure now depends on your agentic system. It moves to the right people, across every system you already run, and is governed at the speed of business the whole way. People, orchestration, governance. Today, discover how the world's largest companies use all three together to achieve AI success. Welcome to DataikuWorld, the AI Success Conference.
Welcome to DataikuWorld, the AI Success Conference. You just saw a great movie, a short film, but let me tell you, it shows AI will transform every organization. But the data says it's a lie. We're living in this world where we think of intelligence as a supermarket where you can purchase a model and AI off the shelf, and that it would be easy because you can just buy it like granola, maybe, and you put it in the basket. But when you look at the data, the data says that almost three quarters of the value of AI is actually captured by 20% of the companies.
I'm not a data scientist, but it tells me that you will have 20% of winners getting most of the value and 80% of losers just getting the scraps of the value of AI. That's what the data tells us today, and it's a bit scary. It's a bit scary because it means that in this world, companies that are purchasing the same thing, the same model, but they will get to a very different level of AI success.
And why? I don't believe it's because of the technology or even about the speed. It's not because one would have chosen Anthropic and the other one OpenAI. It's not because one would have a special deal with Jensen and the other one not. I believe that it's because different companies have a very different way to implement their AI strategy that will lead to success or not.
And companies have a very different way to think about their AI strategy. Even if we all know the same facts about AI, the fact that you have models but ultimately they are becoming more and more black boxes, things that are harder and harder to understand. It won't get easier. We all know that. We all know this fact.
We are living in a world where access to AI could be cut overnight. At least for some models, it happened before. So it will happen again. And we all know that. We are living in a world where AI can solve and do miracles. But in the same world, every time AI is doing a great thing, we are worried that AI is actually potentially stealing data from you or cheating.
Every time AI is doing something great, the day after there is this concern. And we all know that we're living in a world where the same AI that was doing Marvel was also potentially, for a few months, acting on different websites and governments and so forth under the hood. So agents can go rogue, AI can go rogue. So maybe it should slow down. And we all know that. We all know that there are dangers to the usage of AI.
So we know that the model, AI is not the enemy, but the model is not your friend. You have to work with it. We're also living in a world where we know that the cost of AI will change over time and that in order to get the best deal in the supermarket, it won't just be about waiting for a coupon.
You will actually have to go after the market, look for the best deal yourself, decide that you are ready to use different models including open weight, and more and more of you are actually starting to do that. We all know also that it's no longer about model or vendor or even text app to a large extent. The issue we know about AI is the actual return and the ability, the courage in fact, to kill the projects that don't work.
So we all know that. We also all know that in AI, AI is in every boardroom. AI is used by your CEO to make strategic decisions because they use AI instead of other sources to get their information more and more. We all know that. But in the same world, we know that for most companies, getting a proper audit trail of AI, being able to justify decisions with AI, is not possible. So you're making decisions that you can't justify.
The data says that. And this concern, this question of AI, we do know that it's not the model that will solve it for you. It's actually you, probably you in this room, that is in charge of the question of trust in your organization. The model will not solve it for you. It will get harder and harder to get this trust on data and it will become more and more your job, literally your job, to be the guardian of trust in your organization. We know that.
So we all know those facts, and in spite of that, you've got companies that will have very, very different strategies. AI and you will have company A that will go a certain route and implement AI in a certain way, and company B that will go actually the other route and implement AI in another way.
So let's unpack that first. You've got a question about people. You've got a people strategy topic. Some companies decide in fact to not trust their people. They decide that the fastest road to AI, company A will decide that you just bring consultants in, they will solve the problem for me.
Company B instead is like, I think I need to trust my people because they've got the knowledge and the know-how and I need to upskill them so that they can build AI in the long run. And it's a different strategy. You could argue that both strategies may work, but both of those strategies exist on the market today.
You've got an orchestration, a technology model orchestration strategy, which is that some companies, company A, think that AI is the best opportunity to actually rip and replace everything in the organization. Yay, AI is coming! Let's scale everything. Let's put agents instead. When will you be able to do that? Next week? Let's do it now.
And you've got other companies, company B, that would be like, okay, actually in this company we've got some good data sets, we've got some good decision-making, we already have some level of automation, so the actual issue is to mix the old and the new together, to orchestrate things together. And it's a very different strategy. Both of those strategies do exist.
You've got a governance strategy topic then. Governance strategy is like, company A could say, well, compliance, ops, that's great, but let's put that on the side. Let's put that on the side because my best people, they will be slowed down by all of this compliance. And finance and budget people, let's move forward. Some companies do that.
And some other company, company B, would be like, in terms of governance, in fact, if I can see the value and the risk and the cost together, it's actually a way for me to prioritize and actually invest at the right place in AI, knowing where I will get the value, understanding the risk, understanding the cost short-term and long-term. And that's the right way to actually drive efficiency. And it's a different strategy.
Well, we don't believe in the company A strategy. We actually build our product for the company B of the world. The ones that actually want to trust their people, that want to combine together the old and the new, and the ones that think that governance can be a discipline for value realization.
And we believe in that so much that we created this platform not just for one cloud or one data platform or one model, but building it in a way that it can run everywhere, including on-prem, including in hybrid scenarios, over all models, open weight or not. Because we believe that for customers that are the largest companies on the planet, they have this complex setup where they need to be able to apply AI everywhere, and more importantly, to own it in the long run.
And that's the platform we are building at Dataiku. So we keep extending this platform, and today it will be a lot about telling you what we are doing now and what we are planning next. One thing that we're announcing today is these things called agent management. I'm passionate about agent management, maybe because I don't like humans, but the idea of managing agents, I'm like, wow, finally they will actually do it.
And so when you think about it, in the enterprise more and more of your activity will be done in an agentic manner. The ability to see the cost, the value, and the risk in the same control plane and to actually drive efficiency, to decide whether you should invest more in those agents, understand the risk, put people in the business accountable for this, is a key aspect of modern business management.
Modern business management will be more and more about agent management, and that's a capability we want to enable with this platform, so that you can see all of your agents wherever they are and can act on them.
Another thing we are announcing today are extensions of Co-Build. So many of you know Co-Build, it's new capabilities that revolutionize how you can use Dataiku. Because now we build our product in a different manner. We see more and more our product as a gigantic workspace where you can build data and models and agents using Co-Build in order to actually access all of the capabilities on the platform.
Today we are extending it so that anyone in the organization can actually ask a question about data in a controlled manner through Co-Build, among other things that we are extending Co-Build with and announcing today. You'll see today demos of both at the various booths and also during this session, and it's only a fraction of the things we're working on.
Tomorrow in the afternoon in the innovation showcase we'll show what's happening next, including semantic layer, context layer, private AI, on-prem AI, and process automation on data.
I told you this story of the company B's of the world, and so you could not believe me. So I think it's easier to actually show you, which is why today we actually brought on stage three companies that operate with AI differently — Perimeter Canada and Solid — that will actually show how they focus on internal and existing skills, how they orchestrate the old and the new together, and how they actually focus on governance instead of fearing for more agentic slop.
So they'll join today in the three acts of our show: people, orchestration, governance. And so now it's time for the first act, which is people. You got it right. So let's start with this first act, people. And for that I will ask Jed to join me on stage.
Thank you. Thank you very much. Hi everybody. Good morning. Super excited to be here in lovely New York today to talk to you very specifically about a problem that probably all of us have had for, well, since Excel started in the 1990s, something like that. Every enterprise that I talk to right now is sitting on the same problem.
They've been working, as I said, in Excel and Crystal Reports and whatever for 30 years. They have decades of experience, decades of work, decades of information, decades of decisions locked up inside of spreadsheets. Spreadsheets that no one remembers building. Models that maybe ran once and were forgotten. And knowledge that lives in one person's head and one person's computer.
All of this data and expertise can offer deep value to every one of you, but it's locked away right now where no one can reach it. The people who understood the business best, the ones who built these things in Excel, they weren't doing anything wrong per se. They were reacting to the reality on the ground in most major companies, where going through IT to get business information and generate business results is too cumbersome.
So instead you either could write requirements and wait for it, or you could build it yourself in Excel. So everybody did. Right? Today, that's changing.
There's new tools that business users have access to where they could build things even faster than Excel. We have Excel times 10. This problem is happening right now, right? Vibe coding. It's an incredibly tempting form of candy, let's say. Businesses can skip IT completely, and they're not just building spreadsheets anymore.
They're building their own web apps, their own solutions, their own automations, everything on their own inside of Claude Code or Cursor or whatever they can get their hands on, ChatGPT on their phone. This is both an incredible opportunity and also pretty terrifying, I think, especially if you are IT.
So what is the new relationship going to be between business and IT in this world?
At Dataiku, we believe that you unlock value when you both give business experts the tools to build directly and do that under the control and visibility of IT. IT and business have the opportunity to form a new type of relationship. Right now with the advent of AI, it's not business going around IT to use Excel. It's IT providing businesses a safe place to work.
What that means is that when a problem comes up, the business experts can build the solutions themselves as fast as they want to go. But inside the guardrails that IT has set. IT shouldn't need to spend time reviewing every single request for data that comes in. Those days should be over, assessing priority and bandwidth. They are not the blocker anymore to get information back, or they shouldn't be. What IT should be thinking about is how do we set the rules?
How do we ensure that everything built stays inside these rules? And that's how we actually create real business value fast with these new technologies. So obviously that sounds great. Yeah, why not just done, easy? How do you actually do that? How do you give business and IT the same amount of visibility? How do you set the guardrails necessary to do this safely? How do you track where this data is coming from? How do you govern it? What does it even mean to govern AI?
I was talking to a supply chain team at a major manufacturer a couple weeks ago, and they were initially very excited. Somebody had vibe-coded a tax calculator for their shipments, and it was fast. It was providing very positive amounts of information. They felt like they had really made a huge shortcut in their existing process.
The interface looked great. It gave them the numbers and the charts they needed in a much more immediate fashion. They were super excited. It got all the way up to their CFO, and he asked to see the actual calculations underneath this tax and tariff calculator, is what it was.
And nobody could show him. They're like, well, it's somewhere in this 5,000 lines of code. The assumptions were buried deep in there and no one could adequately explain how it was actually working. Has anybody else had this experience or had somebody bring you something that they were very confident in and then you asked them how it worked and they were like, I... raise your hand. Right?
It's becoming an epidemic in the last, let's say, nine months. You could build something that looks so incredible in these tools, but the bottom line is you can't double-click into it and you can't see the data behind it or the assumptions it made. And if something's wrong, it's nearly impossible to identify that it is wrong, or even if it is wrong, unless you're willing to sit there and read 5,000 lines of code, which I know I'm not.
I call these things vibe grenades. So somebody just throws you a little vibe bomb, comes over the fence, and you're like, well, maybe this will explode on me. Maybe this is a really useful piece of information. We don't want vibe grenades. We want to be able to control this process of creation.
So that is why we built Dataiku Co-Build. Dataiku Co-Build is a way where anybody can describe what they want to build in natural language, and code is going to build it for you visually so that anybody can inspect it. That's a huge difference from writing thousands of lines of code to building a visual representation of logic and data.
It can set the guardrails up front inside of Co-Build. Meaning Co-Build is going to go through all of your existing Dataiku connections, all of your existing Dataiku auditing. Everything that you've put in place for analysts is also going to be the case for the agents that you develop and the AI that you develop. You can click into any step inside of the process and see exactly where the data came from, what it's connected to, and nothing is going to ship to production without human sign-off.
Very happy to say that Co-Build is generally available today. We released it in mid-June. We've been rolling out updates to it basically daily. I don't sleep anymore. Customers are loving it though, which obviously feels great to build something that customers love. It's our fastest adopted product ever at Dataiku. We already have thousands of users on it. Close to a quarter of our customers are already running it live on their production data.
Thank you for that. Thank you for adopting it and running with it so fast. It's excellent. And the greatest thing is that customers keep telling me that it's changing the way they use Dataiku and changing the way they think about building with agents and building with AI in a profound way.
Since we released it, I personally don't really click anymore. I just talk to it to use Dataiku. And I know Dataiku as well as anybody. I think I've probably spent more hours in the Dataiku application than almost anybody on earth, and I still would prefer to talk to it. It's that powerful.
Really, really encourage you to check it out upstairs. There's great demos for it. We'll actually see a bit in a minute. And also just open it up in Dataiku. Give it a shot. It's amazing.
So with that, I'm really excited to announce that Co-Build is expanding into every corner of Dataiku. Co-Build started inside the project. You open a project and then a data scientist or a data analyst talks to Co-Build and it builds a flow for you. But the first place that we're expanding it is Co-Build Insights.
So Co-Build Insights is not just for your classic data scientist or data analyst. With Co-Build Insights, anybody can log into the Dataiku application as a whole and ask very general questions, and it's going to find the right projects, data sets, information through our semantic layer to provide the response back. So this really opens it up to a much less technical level of business user. Really exciting. Going to be available within the next few weeks.
Co-Build Agent Management. You're going to learn more about agent management in a bit, but we have also injected Co-Build into every aspect of agent management. So as you start tracking new and additional agents across your platform, across your organization, you're going to be able to have a conversation with Co-Build to understand exactly what all of those agents are doing, to fix them, to bring them up, to bring them down, to ensure the topics those agents are talking about are correct. Essentially, you still have this conversational input not just to build agents but about all of the agents across your entire estate inside or outside of Dataiku.
And then finally, and this one is near and dear to my heart, probably the one thing I've been working on most closely personally for the last four months, is Co-Build Headless. Now Headless is very interesting. We recognize that people love ChatGPT and people love Claude Code. Who here uses Chat? Who here uses Claude Code? Okay. Who here uses Cursor?
Yeah, like these things are great. It's a great interface. I use it constantly. I love it.
Headless is a plugin that you can download from the ChatGPT store, from the Claude Code store, that you can install onto your own custom coding harness and begin interacting with Dataiku.
It's going to, underneath the covers, leverage code to build whatever you want inside of Dataiku. There's a couple advantages to this beyond just people liking ChatGPT. One is that you can mix and match other third-party services very easily with Dataiku.
Quick example. The other day somebody had a question about one of our clients. They're asking about some usage data and I pulled up Headless and I said, Hey, Headless, look up the usage data around this client and then set a calendar invite to set a meeting about this, and then also create a report and put it into Drive and send it to the account executive for the client. All in two sentences. And it combined the Dataiku plugin with the Google Drive plugin with the Google Calendar plugin with Slack. Put them all together in a few sentences and I was able to work across all these things. Very, very powerful.
The other part is that it makes it incredibly easy to migrate workflows into Dataiku. So if you have something sitting in Excel or sitting in Alteryx or sitting in SAS, now I can take that information, say, hey, Chat, look at this, transform it into a Dataiku flow, and off we go. And we're going to see that also in a minute.
So these three things I think really advance Co-Build in super broad directions in a very powerful way. It lets you feel like using Dataiku is right at the tip of the spear as far as technical capability. We do not want to be a blocker. We don't want to hold you back. We just want to make sure you're safe and powerful.
Now I want to hear from a customer who's been with Dataiku for five years. I'm really excited for them to be joining. This is Kyle Benning. He's the CEO from Purdue Farms, here today to chat through his experience and how he equipped his team with the tools to build all across his organization. Kyle, come on out here for us.
Thanks very much. Sorry about that. So Kyle, there we go. Kyle, five years ago, you made a bet on Dataiku and you made a bet on bringing the people closest to the business to let them do the building themselves. What made you willing to make that bet five years ago?
When I walked into the company, I was the first data person, the only data person. So it was either build out a big function or bet on the business. It was either try to sell the executives that I was smarter than the rest of their decision people or just enable their decision people. So it was kind of a no-brainer to say the way to succeed is to enable the existing decision network and give them a tool that teaches them how to do it a new way. So they come out of just doing data and they actually have a way to structure it, and then build a team that does that operating model and sets everything up for structure. So we were able to control all of those decision networks that already happened.
What was the immediate impact of that? What were your early wins that you saw?
What we did is a lot of little quick low-hanging fruit stuff, but in three years we did over $35 million worth of just small productivity yield, all the bottom-line savings. But it let us build out our practice while saving money.
Brilliant. Brilliant. And now that you've expanded to many business users across a lot of different teams, over $100 million in value, I think how has the structure, the way you support all of them, changed being the central team?
It's actually just scaled. When we started, we started with two or three teams. We had one what we call the player coach, which was really data translators, people have used. It was really what I call my mini coaches. They teach those function teams how to do data and make data work.
What we do is we pair them across multiple functions that have data. So now they're the connecting tissue that connects those teams together and gets them to be collaborative. And we just scaled out from there. So as the teams grew, we added more player coaches, we added more engineers, but it all hinges on the same operating model and it's just scale.
Can you tell me a little bit about the ecosystem stack that your team is built on? Beyond just Dataiku.
Very simply, Snowflake, Dataiku, and Azure.
So and Snowflake is, you know, the processing, storage where some of the technical guys will play, but the business orchestrates everything through Dataiku and that controls our whole environment.
Brilliant. Brilliant. So what about Gen AI? Is that happening yet? Where is that coming from, going to? Tell me what's going on there.
I think we've begun the journey with Gen AI. We're starting to play with it just like everybody is. We've really taken, because we're going on the business and the business is working through process transformation, we've pumped the brakes on, hey, throw a whole bunch of agents out there. Go do agents. Go do agents because they don't really know what they're identifying. So we first start them with, let's start with what is the business transformation that you're trying to do. People ask me, what is our AI strategy? I always say we don't have an AI strategy and we never will. We have a business strategy that AI is enabling.
I love that. Yeah. You're not doing it for its own sake.
We're not doing it for its own sake. I have to be smart enough. My team has to know what AI should do, and we have a little bit of an AI strategy so we know where to tease the business on where to go. But it's all about the business strategy and working on those processes and making sure that we're enabling and identifying the things that make life better.
Makes a lot of sense. Makes a lot of sense. Looking at the future, where do you see all of this going? What is Purdue's data teams? What are they using? What does it look like in two years?
Well, the thing that's exciting me is Co-Build, quite honestly. Let's say we had AI first with visual recipes, right? And so all of the people who were early adopters and quick followers, they took the visual recipes and ran with it. Now that we get Co-Build, we get all of the laggers who have been denying, they can now just start talking with their data and doing the work that they do anyway through their conversations. But we just enable it that way.
Very, very cool. Well, thank you so much, Kyle. It's great to hear from somebody who's actually doing this and we really appreciate you and everything you've done.
Purdue Farms has been in business for over 100 years. We're a 10-plus billion dollar company. We do food and agricultural products, protein and grains. We have over 3,000 farmers that we partner with. Five years ago when I started at Purdue, it was a very ad hoc relationship with data and it really took a long time for them to get to an insight.
The data foundation is going to be critical for any AI success, whether it is leveraging analytics today or building AI tomorrow. That's kind of the crux of why we are successful, is because Dataiku gives us the place where the business can just use data and they don't have to worry about coding. It does it.
There's usually a tug of war between IT and business. There's always that question of how much freedom to give to people.
We actually said, hey, let's lead with the business. They log in and they start playing with their data directly. When you have the guardrails already built into the process, it's just the process.
We don't view governance as a hindrance or a constraint. We use governance as an enabler, really as an accelerator.
The associate-first ethos at Purdue is really about empowering the end user to do that.
Our role here at Purdue is to make sure that pack of chicken is on the shelf when a consumer walks by. Transportation costs are quite high these days. We have an issue with shipping air. And so we took information from multiple sources, put it all together to identify an opportunity worth millions of dollars.
If those boxes aren't 100% full and there's air at the top of those boxes, we're shipping empty space. So you're able to reduce our shipping cost.
I've written SQL by hand my whole career. Even as an IT professional, I prefer to build my data pipelines in Dataiku. One of the things I really like about the way that Dataiku has enabled AI is that you can build your pipelines with Co-Build and then see those pipelines, confirm that's actually what you want.
Data and Co-Build and Snowflake working together is really a cool collaborative set. Snowflake is that compute and that storage layer.
A strong data foundation is very important and Dataiku has enabled us to build a layer that is trusted, governed, and consumable by different business functions.
To be able to be successful in AI, we need to have a full-scale platform that does everything from the grassroots little things that we did for our first three, three and a half years of the program, saving a total over the last five years of over $100 million of value, leveraging Dataiku to make the business better.
I've never seen another tool that enables business users to get in, work with their data, understand it, build pipelines, and get their answers.
So we take the processes and bring them into the next century. AI is going to be a key cornerstone of that, and Dataiku will be the key cornerstone that helps us enable those processes.
Thank you again to Purdue for sharing that great story with us. I would like to next bring out Cassie Schulz. Nailed it. All right. A senior solutions engineer here at Dataiku. Worked with her for years. She's brilliant and she's going to be with us all morning. Actually, she's here to show us exactly what Dataiku looks like in action. To start, we're going to be talking about Fieldstone Foods. Fieldstone Foods ships chicken. Might sound familiar to someone you just met. But today she'll be playing a transportation planner at Fieldstone Foods. Let's give her a warm welcome, which we just did, and let's see what we're doing here.
Hello everyone. So, I'm here today because one of my planners, Bob, is about to retire after 20 years at Fieldstone Foods. Bob's got this crazy Excel workbook. It figures out how we can best package and ship our chicken across the country with the minimum number of trucks.
We're going to wiggle this around real quick. Cass, Bob's the only one that knows how that works.
Bob loves us how this works.
Any news on this? Give us one second, please.
Well, as we're working that out, the problem that we have here is that Bob has built this Excel workbook that I spoke about before. And this Excel workbook is actually critical to the operation of the entire company. How many people in here have an Excel workbook that maybe you're happy about, maybe you're not happy about, that your company relies on, or a hundred of them, or 500 of them?
What's the maximum number of tabs that we've got in one of those workers? How many of you have multiple ones that call each other in some horrific fashion? Point being that this type of work exists everywhere, and we have this new opportunity right now to transform and evolve this work.
It was always very, very heavy, probably too heavy to move from Excel to anything else. It would have taken hundreds or thousands of man-hours. All of a sudden we now have the capability to understand. All right.
This was all part of the actual talk. So we're good. All of a sudden, we have the capability right now to look at a workbook like this and really dive into it without having to manually look at thousands of formulas. So, Cassie, what are we looking at here?
Well, this is Bob's workbook and my team can run it, but we can't build on it because no one but Bob understands what's going on underneath the hood. So, let's try and make that something that other people can use and build on.
Great. How do we go about doing that?
Well, I'm gonna ask Headless if we can help out.
Looks like we're just running that in ChatGPT. So, she's asking migrate this to Dataiku. And she's going to find that Excel workbook, pop it in there, and off we go. So, it found Headless. It knows how to connect to my instance. It's now, you can see that it's asking me questions or asking Cassie questions about what that workbook is doing. It's going to understand the underlying components here and build that out as well as looking at the validations necessary to build it into Dataiku. So it's verifying that that information is the same. But we don't have to just trust ChatGPT plus Dataiku Headless. We are going to dive now into the project and see exactly what's happening. Cassie, tell me more.
So, at first glance, this looks like it's still kind of a lot. But if we look a little closer, we'll see that Headless has documented every single step along the way from the input data sources to the data transformations to explain to anyone who comes into Dataiku what was done and why.
And if we look a little closer, we don't just have to trust this. We can actually leverage the validation steps that Headless put into play with Co-Build in order to make sure that across all 72 products that Bob was responsible for reviewing, we've got every single metric matching between that workbook and Dataiku. Looks like it's a perfect match.
Great. What more could we do, though? I feel like just because I've replicated Excel here, can I take it farther? Let's go ahead and run one thing that comes to mind, which is every single review cycle, Bob has to go into source data, extract some data, come back into Excel, import it in, run the Excel workbook, fix anything that goes wrong, and there's a lot of room for error along the way. So, why don't we save our team all of that time?
That makes sense. I hear this a thousand times over and over. Oh, yeah, I go to SAP and I extract the spreadsheet from the ERP and then I put it in Excel and then I work on it for a week and then I give the information back. What if we could just connect directly to the underlying data and automate this?
So we can take this even a step further because in addition to having to run and pull that live data every month, Bob also has to run the workbook manually. Why don't we just automate the entire thing?
I'm going to ask Co-Build in a new thread to automate this pipeline and send a Slack to the packaging team when it's done. Let's make sure I spell Slack right.
So, while this is running, Co-Build doesn't just do automation and recipe replacement, right? I could also have it build a machine learning model or an agent or whatever other fancier Dataiku things that might be up my alley. Is that correct?
Exactly. So, we're just going to start at kind of the basics here. Let's get this project moved into Dataiku. Let's get it automated and not dependent on static data sources. And then the next level is how do we build upon it so that more people can consume it or we can use it in different and advanced ways.
Okay. So you might be thinking beyond this and we'll make sure that this runs every day. Now our team is going to have a copy-and-paste replacement from that copy-and-paste mechanism every day. It'll check for new data showing up in their books.
Love it. But what about folks outside of your specific team? What I can imagine a higher-level executive maybe wants to come in and understand what's going on overall with my logistics and my packaging. Can I get that type of information now that this is here instead of having to go and ask Bob about his Excel file?
That's probably how I would have done it before, right?
Yeah. And that's what Co-Build Insights is for. So, there are going to be a lot more people than just the builders who are involved in a project like this. Whether it's someone who's involved on the business side and has questions or stakeholders that want to understand the implications of the projects that are being done, anyone can come in and ask a question like, where should Fieldstone focus its packaging review first and how much transportation opportunity is at stake.
Wow. So just because I was able to move this project into Dataiku, I'm now able to ask these very high-level questions about it with no additional work and get the information back that I might need as a high-level executive.
Exactly. And so under the hood, Co-Build Insights is leveraging all of the governed data sources that we've put into play here, and we'll get an answer, which is that we can potentially save 11 trips over the next four weeks, which is 38% of the network total, if we focus on the Dallas to San Antonio DC route, and then we'll get some more metrics as to what we can prioritize next.
That's really terrific. Thanks Cassie. I have one more question though. We did this as Excel. We made it look pretty easy. Is it Excel only? Is that the only place Headless can work or could I do it across other types of legacy systems?
Definitely Excel is not the only place. This is just one of many common legacy systems where institutional knowledge has been baked into the processes that have been built over lots of time. And so when we think about our team particularly, we're looking at Alteryx workflows. We're looking at SAS and we're seeing that there are a number of different opportunities and ways that we can bring that into Dataiku, open it up for the wider public to view.
Great. This looks like a fantastic new future for Fieldstone Foods. I think we're going to be a lot better off. Thank you so much, Cassie. Let's give a round of applause.
So, what's the takeaway here? For me, it's that when the people who understand the business are given the tools and the trust to build new technology quickly, the value that's been sitting locked away for years across a variety of different programs, variety of different people, laptops, applications, starts showing up, starts being actionable, starts being successful. The businesses that succeed right now are going to accept these new builders. And with Co-Build and Dataiku, you can finally feel safe to unleash them and succeed in the future. Thank you so much.
I'm now super happy to announce Clay Stanak for the next stage of our conversation.
Thank you. Thank you, Jed. These were great insights about who builds things and how. But work doesn't end with building. How does everything that gets built fit together? Where do these things live? Where do they get their information from? Where is intelligence? And who owns it?
All of you, every company in this room, most probably already has a vast portfolio of AI building blocks, but usually it's very fragmented. You may have a model over here, an agent over there, maybe a data pipeline that nobody remembers building, and usually they don't talk to each other.
And the thing is that an agent that handles one task usually doesn't move the needle on its own because that task is only one part of a bigger business process. And if you just put an agent that handles one task, the business process still keeps running around exactly like it did before.
So, in order to change the way your business works, you usually already have the pieces you need, but you're missing the ability to weave them into something that runs the way your business actually runs and that gets real value out of things that you may have already built.
And we designed Dataiku exactly to solve this challenge, to meet you wherever you are and to keep growing with you as the technology evolves. You can connect whatever cloud data platform, database or LLM that you are using, deploy however you need and add or swap models, vendors or platform later without having to rebuild everything you did. This is really what interoperability is about. You have some stacks but now they can work together.
And as we said, most real business processes are too complex for a single agent to run alone. So that brings together multiple agents working alongside your models but most importantly working alongside your people who stay in control of the decisions that truly matter. We call this capability structured visual agents.
And for them to actually run your business, there are three key ingredients that you need. The first is that they have to carry judgment the way your best people do. The second is that they have to understand your data and your business-specific terminology. And the third is that they need to adapt quickly since everything around them is changing.
Let's start with judgment. A structured visual agent is built as a flow of blocks each with one clear job. It can be calling a tool. It can be ingesting data or it can be handing out a task to a specialized agent built exactly for that piece of work. And some blocks can run alongside together and bring their results back together. And the flow of the blocks carries context so that any decision made three steps later still remembers everything that came before it.
And so this is how judgment gets built into the system. Some of it is encoded...
Once with fully deterministic decision making. Some of it uses the full power of agentic and the rest stays human. You as builders decide where a person needs to make the call. In practice, real enterprises benefit from variety. You have different models for different jobs. You have different teams for vendors for different teams and the people making the call that matter. And structured visual agents make all of these work as one coordinating system that connects your data, your models and judgment into agents that can really drive the business forward and deliver real business value.
The second ingredient is business context. Your agents need to understand your business terminology and specific industry nuances. And that's why we added a separate layer to that called semantic models. For example, let's say that you use internally the term active customer. But what is for you an active customer? A semantic model defines the entities in your data, their relationships, the business metrics and glossary terms that you use and optionally even the actual code needed to compute them exactly, for example as SQL code. So that active customers means each time the things that it means for your business. And then once you have defined that, the structured visual agents can leverage this semantic model so that when the agent needs to compute okay how many active customers were here last quarter, the query that gets executed in your underlying data is the right one grounded in the terms that your business uses.
But after judgment and business context, if you want success with agents, adaptability is probably the most important ingredient. The market today is in constant motion. New models arrive, regulations shift, providers get pulled, and your teams build on different platforms. This is very simply the reality, and without adaptability, your agents will die. Dataiku's LLM mesh sits in the middle of structured visual agents so that you can swap a model or even a whole provider without having to rebuild the agents and the layers on top. So that when change happens in the outside world, your agentic process continue to thrive even though the technology underneath them keeps shifting.
And I'm excited to show you how our next guest have done this. Building complex agents and orchestrating AI across more than 18 teams. Let's hear from Air Canada.
