Cofounder, Dataiku
Florian Douetteau, CEO at Dataiku | Live from HumanX 2026
Send us Fan Mail (https://www.buzzsprout.com/520474/fan...)
Florian Douetteau is the CEO and co-founder of Dataiku, the enterprise AI platform he has been building for over a decade to make data and AI accessible at scale. Recorded live from the floor of HumanX 2026, this lightning round explores how the rise of agents is reshaping the role of data teams and what a genuinely AI-native way of working looks like in practice.
Florian and host Dan Turchin go deep on where agentic AI actually creates economic value, why most enterprises are still thinking about agents the way people thought ab...
About Florian Douetteau
Florian Douetteau, co-founder and CEO of Dataiku, has been discussing the evolution of enterprise AI and the challenges posed by agentic systems. In a July 2026 interview, he said that large language models (LLMs) are becoming commoditized, with the key drivers being inference speed and energy cost. He argued that while AI can automate many tasks, it cannot solve the problem of accountability, as agentic systems concentrate risk and responsibility rather than distributing it. Douetteau also expressed skepticism that "vibe coding" will be the final interface between humans and machines, predicting a shift toward visual or domain-specific languages to manage human-LLM interactions. He warned that as the cost of creation approaches zero, the cost of managing complexity will rise sharply, creating a risk of organizational chaos in large enterprises. In earlier appearances, Douetteau focused on the need for governance and data quality in AI adoption. He described "semi-deterministic" agents—such as those used in procurement or marketing budget optimization—as a promising category for enterprise use, contrasting them with open-ended project management agents. He noted that many enterprises are still taking "baby steps" with agents, with many so-called agents being "glorified assistants" rather than true autonomous systems. Douetteau emphasized that giving more people orchestration powers requires stronger governance, and that the data products that actually matter in business are often "shadow data products" like Excel spreadsheets with VBA. He also spoke about the societal implications of AI, stating that the outcome could be dystopian if it leads to widespread replacement of workers, or positive if it accelerates research and economic efficiency, and that the erosion of trust in people's ability to make a living could lead to social unrest.
Transcript (28 segments)
D
Dan Turchin0:00
Yo, it's the future. Don't hit the snooze. AIs are giving humans the boost. Team humans rolling. Brains in the game with Dan Turchin. Y'all remember the name. People ringing the mix, no stopping the flow. Better work, better life. Yeah, we're ready to go. Robots and humans are collab, you know. Okay, let's start the show.
Florian, you need no introduction, but for the sake of the podcast, we're here at Human X recording this live. Could you share a little bit about your background and maybe the founding story of Dataiku?
F
Florian Douetteau0:31
Yeah. So, I'm Florian, the CEO and co-founder of Dataiku. It's a company that I created more than a decade ago. My background is that I wanted to change data science. And in the early days of AI in the enterprise, I wanted to make sure that it was super easy for every enterprise to scale data and AI. And so, we built over the years this comprehensive platform where you can do everything data and AI. Prepare the data, build models, and now build agents so that enterprise can actually make it their own.
D
Dan Turchin1:00
All data platforms have had to reinvent themselves in the last three years or so. I've been following the Dataiku journey for long over a decade. I'm curious how you've adapted the company to an AI-first world.
F
Florian Douetteau1:14
Yeah. Last year, we added robust agent capabilities in our platform, as in a full agent studio with the ability to create different types of agents, and agent testing and validation and guardrails, like all of the things. And finally, it changed the way our customers really used the platform. Meaning, the data scientists, the data people, they were using the platform to build models and a small app on top and so forth. And it's been a decade. Data, ML, data science. And the same people started to build agents. First, small ones. Instead of stopping at an app, can you automate this or that? Can you make the interface of the discovery of the information conversational for your own internal customers? Can you do the next step of action taking coming from the model and automate all of that? And so this agent framework just changed the consumption layer of our platform, making an agent one of the objects, the main object that customers are creating with the platform now. Which is, yeah, an interesting change. It's the move from apps to agents that we happen to see in real time in our own platform.
D
Dan Turchin2:25
The old tools of the trade for a data scientist were a data platform like Dataiku and analysis and things like R or Python. And the skill was they would build and own the model and then typically hand it over to be operationalized by someone else. A developer, an app team, etc. What do you see changing in terms of the org structure? Is all that getting compressed into fewer roles?
F
Florian Douetteau2:49
I think that is a model of a no-one-spoke model where the business logic and all of this orchestration and understanding of the data needs to be pushed on more and more to the business teams. Because what we're talking from a future of work perspective is the fact that business teams need to automate themselves in a nutshell. Because they do have the understanding of the business logic, the understanding of the trade-offs. Meaning it's not true that you can just prompt your way and put all of your data in code and say do my job. Doesn't really work like that. And those business teams need to become those enablers. Need to become able to do AI by themselves. The data scientist may stand more in the background making sure that the technology is safe, that it can scale, and so forth. So it's a change of role for them. That's a first thing. A second key thing is the way you build data pipelines, AI pipelines is changing very rapidly. In our platform already, you've got lots of AI assistants and you actually build pipelines on the fly. And our platform is now more and more kind of like a canvas for an agent to build the pipeline for you instead of drag and dropping into the interface. And so, I think those two changes are changing the role of the data scientist or the ML people or the data people at large. One, because business people need to be enabled by themselves more and more. And second, because the type of productivity you can get with data is being demultiplied. So, those two factors together will and are changing already the role of the data people.
D
Dan Turchin4:27
The best data scientists are great storytellers. They're really telling stories. And their medium is data. AI now wants to be the storyteller. Are data scientists able to tell better stories with AI, or can everyone now be a data scientist because they're partnering with an AI storyteller? How do you see the storytelling aspect changing?
F
Florian Douetteau4:50
Yeah, I think that's, if we take a step back, storytelling is part of a craft you needed to build as part of decision-making in the enterprise. You're telling a story in order to influence people so that they take a decision in practice because of that. With AI, because of the automation you can make, the process itself, the ideal process is changing. So, example, let's make it concrete. The old way of marketing and campaign management is you ask a data scientist to tell you something about your audience and the targets. The data scientist is doing data storytelling, and the business stakeholders take decision in terms of crafting their campaign or budget allocation of the marketing budget, and so on. Based on this storytelling. That's the old way. The new way of marketing, because of AI and the type of more granular automation you can make, is that you're listening more constantly to existing signals. You get way more personalized on an individual per individual basis like this type of trends. You can do it because you can identify your way to mass creation of content. And so in that particular instance, the need of storytelling is disappearing itself. Because the type of decisions you're making are way more granular and are happening differently. So indeed we have to understand that to some extent part of the role of the storyteller or part of the storytelling will be disappearing. Which is maybe sad. I mean I like stories in general.
D
Dan Turchin6:19
It was probably even less than five years ago when the notion of being able to speak to your data using natural language, it was science fiction. I mean you and I both go back to a time when NLP, natural language processing, it was a brute force activity. I mean it was very crude. Very hand, you know, you had to hand manage the data. And now we get that for free essentially. Does that just expand the surface area of the kinds of questions we can ask? I mean it's such an unlock but it also changes your business model. It changes how businesses operate. How do you think about the blast radius of being able to speak to your data?
F
Florian Douetteau7:17
Yeah, I think that speaking to your data is becoming a capability of the data platform. Meaning Databricks and Snowflake, you can speak to your data. And maybe that's why ultimately you should buy Databricks and Snowflake to have a free speak to your data. And it's now less about BI for that but directly you can do that directly on top of the data platforms. Now what you do with our platform, for instance, is you've got use cases where you want just to enable people in the business to speak to their data, to get point answers to some point questions. You've got other use cases where you want people in the business to build repeatable scenarios of planning and something they can do over time and on which you build a business process. And for that, you need some consistency. You still need to build repeatable pipelines or build an agent that can be fairly deterministic in terms of how it gets the data and so forth. And that's where our platform now it's new. It's no longer used to speak to your data because anyway, you can speak to your data with whatever data platform and so forth. So, it's kind of like a shift to the market. To some extent, yeah, BI itself is maybe not disappearing, but definitely is changing in terms of scope.
D
Dan Turchin8:37
But Databricks and Snowflake also want to be the agentic layer, right?
F
Florian Douetteau8:41
Yeah, but like being the agentic layer, talking about agents now, we're talking about agents in this current period now in '26 very naively, and I say that for any listener of the future because in '26, when you talk about agents, you're like people talking about websites in '97 with e-commerce or blogs not having been invented, that concept, and you're just doing HTML with no Ajax, like you're all just doing very naive stuff, not understanding what you're doing. That's what we're doing with agents today. And like everything will be agents, so yeah, if you do software stuff, you're doing agents. The question is from the perspective of the enterprise, what type of agents are built on top of which platforms? Yeah, your customer support agents are probably built on top of like very specialized platforms. Everything which can get or talk to your data is probably directly done on your data platform. At Dataiku, we're focusing in fact on lots of agentic use cases that are not token intensive. And very boring in the sense of being a back office decision heavy type of task. Supply chain inventory management, predictive asset management in a factory, clinical trial optimization, where you use agentic to optimize, you have a mix of heavy usage of the data. You are using tokens of course but like not so much compared to code or whatsoever. And what you focus on is the predictability and the soundness of the decision making. And the way I see it in the enterprise, you need whatever happens a dedicated platform for that because you need people in the business to be able to open up the thing and understand how it works. Because it's decision making, you can't just be black box on that because the core of the enterprise is to make decisions you understand that you can repeat in the future.
D
Dan Turchin10:40
Because you have a background in philosophy and because this podcast is AI and the future of work, I get to ask you this question. When humans start to interact with machines that can do things that only humans were ever supposed to do, how does that change our relationship with work and what does it look like if you play this out a decade, what's the work that humans are going to be doing?
F
Florian Douetteau11:06
And be honest and serious, I don't know. Meaning I still wonder. Something that I can see more short term, which I think can be a first step or an indication is that when I think for instance on the role of data people now, so what we are building and doing with our platform, it looks like you're leaving work at 5:00 or 6:00, you collected everything you wanted the system to do through the various meetings of the day and some recommendation. The system is running at night doing lots of stuff, building lots of pipelines, analytics. You get back to work at 9:00. You review all of that. You connect things together. You take a step back. You're like scratching your head. You're like, 'Yeah, maybe you should do that. This thing makes no sense.' You exercise judgment. At 11:00 you tell the system to relaunch itself based on all of those iterations. You go for lunch. If you're in France, you go for lunch from like 11:00 to 1:00 p.m. Giving two hours for the GPUs to actually do stuff. That's maybe when France would actually have an edge on the market because of our ability to actually stop working. Well, the GPUs are working. And so then at 1:00 to 2:00 p.m. you get back. The system has work and so you have to understand it will be a way of working where asynchronously you've got other systems doing work for you and your job is becoming about reading and giving feedback to them more asynchronously and like it's a different way to work. I'm not even getting into the bottom of like what will be a job versus not a job. I mean there will be a different type of interaction or way to actually work. Less about navigating on a computer and like things. It will be a different way to actually interact and work. I think that's a step I can see happening very soon, in fact. In a decade seems to be very very far away compared to this what I'm describing, which I think could be the reality in two years, quite literally.
D
Dan Turchin13:13
What you just described sounds utopian. And yet what we read in the popular press makes it sound dystopian. That the machines are out to get us. It's about what AI is doing to us rather than for us or with us. How do you think about human agency given that scenario that you just described?
F
Florian Douetteau13:36
The way I think about it is though again, I think about it about authentic use cases. A part of the reason of the dystopian is that there is lots of hype or momentum towards those very token heavy use cases such as automating developers or automating call centers on customer support. Yeah, lots of tokens there. And there are use cases where you replace humans and it's not that clear that you will generate incremental pure economic value. I mean, not so clear that you generate growth out of it. As in new stuff being done better kind of. You know, you can see replacement only. A thing that you need to balance that with use cases in the enterprise where you actually bring efficiency. As in the economy itself is working better. I can get more out of my factory because of better maintenance without having the need of more materials or more energy. I manage better my inventory and supply so that I've got less waste. I'm better as a bank even at giving credits in one day instead of one year and it helps the economy to go faster. You need those use cases also to work so that you can bring economic growth and meaning to AI. I mean, it's dystopian if the only thing we get of AI is some people being replaced and asking for a salary. It's positive if you can accelerate research, if you can make the economy actually more efficient and so on. And so you have to balance those use cases in order to get from a dystopian to utopia. Or at least something we can cope with.
D
Dan Turchin15:16
Doesn't that describe the historical relationship of humans and technology? It's always been disinflationary in the sense that we can produce the same output with less input or more output per token of input.
F
Florian Douetteau15:31
You are also with technology that is also waves where you create brand new products and brand new services. And I guess there is some indeed economic incentives or moments where if you create too much of the things that are destroying or too fast compared to the things that are creating something new, indeed you create an imbalance and it's a problem. It's a problem meaning that people are not happy about their work, are not happy about the balance, and ultimately society is based on the contract that people have the trust that they can have a chance of making their living. And that's why they give to the states the monopoly of violence. If you remove this trust, the monopoly of violence will not just be by the states. Meaning people will revolt. That's the actual thing that will be happening.
D
Dan Turchin16:26
In the 18th century the Industrial Revolution was catalyzed because what was a machine and what was a human was very obvious and the machines, it was clear what they were doing versus what the humans were doing. I would argue this is different because the machines feel a lot more human-like. Does that mean is there a relationship with machines changing because they seem like humans?
F
Florian Douetteau16:57
I think it is. I think the main, but to me the main difference with the Industrial Revolution is that the Industrial Revolution happened in countries that were also able at the same time to build empires and colonize either the world or the rest of the American continents in order to fuel the industrial revolution with more land and more perspectives. If you actually are building this type of technology revolution but like being in a fixed state in a narrow mind where you don't have this expansion potential, indeed you create more stress. Meaning industrial revolution or agricultural revolution was also what accelerated decades of communism in China and Russia. So, depending on the state you can actually have very different outcomes as a society. I think that's the main difference. Then indeed as individuals, it does change what we mean by intelligence, whether it is a machine. It has I think the deeper meaning is that if you've got another intelligence that's human, it changes the meaning of what it means to be a human because for hundreds of years we've defined humans as being in fact the person being intelligent. And that will no longer be the case.
D
Dan Turchin18:21
Got to get you off the hot seat because we're going to run out of time. We're going to lose our access to this great podcast studio, but we're going to pick that topic up another time. But before I can let you off the hot seat, I've got one last important question for you. Let's say that in a decade, long time from now, in a decade, you get half of your time back. What is Florian the human going to do with the 50% of your time that is no longer spent doing some of what consumes your day today?
F
Florian Douetteau18:56
Hopefully it would be talking to people. When I look backward I think that the advent of technology made me less actually talking as in like speaking and taking quality time in person with real humans and maybe having time back will rejuvenate a bit of that.
D
Dan Turchin19:20
So you're saying in the future we're going to have more time for this? To talk about the nature of humans and our relationship to the universe and what is intelligence really?
F
Florian Douetteau19:32
Yeah. I think that could be a, if the topic is right or the discussion is right or agreeable, I think it's a nice future.
D
Dan Turchin19:47
Well, I want you to come back frequently and have more versions of the conversation about that future. The world needs us to have more of this conversation, right? Thanks for hanging out. Great to meet you and that's all the time we have for now on AI and the future of work. As always, I'm your host Dan Turchin from PeopleRain. And of course, we're back next week with another fascinating guest.
F
Florian Douetteau19:56
Thank you.
D
Dan Turchin20:09
Yo, it's the end of the show. Time to power it down. AI's the assist, but humans wear the crown. Team human in the house, got the vibe so clever. Dan Turchin from PeopleRain making us better. We learned about AI for humans. We're ready to conquer the future. That's all for now. See you next time.