Florian Douetteau and the trouble with brilliant machines

Florian Douetteau has a useful story for anyone who has fallen in love with an AI demonstration. At Dataiku, his own company, an internal application involving sales calls came with an unwelcome discovery: its projected running cost was around $100,000. The number stopped the conversation. “$100K? No way.” Optimization suddenly looked rather more interesting than another demonstration.

He told the story in 2024, when the prospect of putting generative AI into everyday business was producing considerable excitement. Here was the co-founder of an AI software company describing the moment he declined to pay for quite so much intelligence. The anecdote has the virtue of an invoice. It gives an abstract argument about control a number that an executive can understand.

It also offers a way into Douetteau’s career. He is interested in making sophisticated technology usable by people with ordinary constraints: budgets, colleagues, deadlines, existing systems. A machine can be brilliant and still be an awkward employee. Someone must decide what it costs, what it is allowed to do, and whether its work helps anybody.

A detour through the search box

His route into that problem began well before the current fascination with chatbots. At the École normale supérieure in Paris, Douetteau was headed toward research in formal logic and type systems. In 2000, at about twenty, he joined the startup world. Exalead, the French search-engine company, became his early professional education.

Search is a useful place to lose one’s innocence about information. The user sees a box and a list of answers. The engineer must contend with the machinery that makes those answers possible. Douetteau worked across engineering, product and data, accumulating experience in the distance between an elegant technical idea and something another person can use.

He later worked at social-gaming company IsCool Entertainment and consulted for Criteo. Those stops brought him closer to commercial questions about how people behave and how businesses make decisions. By 2013, he was building Dataiku in Paris with Clément Stenac, Thomas Cabrol and Marc Batty. Their proposition gave technical specialists and business colleagues a shared place to work with data.

Three dialects, one conversation

The founding problem was cultural as much as computational. Statistics, software development and business expertise came with different habits and vocabularies. Douetteau wanted those groups to work together. Even the company’s name made an unlikely introduction: data meeting haiku, the Japanese poetic form. A spreadsheet had acquired a literary acquaintance.

The distinction matters when considering his public ambition to make AI more widely accessible. Giving someone a tool is only the first step. A person who knows a business process may need to examine the data behind a prediction. A programmer may need to understand why a result matters. Their ability to question one another is part of the product’s usefulness.

In his public biography, Douetteau describes an ambition for AI to become an everyday instrument through collaboration across the company. That ambition resists the idea that one specialist department can simply deliver intelligence to everyone else. It asks the people doing the work to participate in deciding how the work changes.

An ocean was part of the design

Dataiku’s American expansion belonged to its early thinking. The company opened a New York office in 2015; by the following year, Douetteau was dividing his time across the Atlantic. Building an international company required a human bridge between teams as well as a product that could travel.

That history helps explain the circle around him. Stenac, the technical co-founder, came from Exalead. Investor Matt Turck and FirstMark became part of Dataiku’s journey, with FirstMark leading its Series A in 2016. Douetteau has returned to conversations with Turck over several years, discussing both the software and the practical work of running the business.

The financial milestones arrived in a different register. In December 2022, Dataiku announced a $200 million Series F led by Wellington Management at a $3.7 billion valuation. The announcement placed a large number beside a company whose daily proposition remained unusually workmanlike: help organizations turn their data into applications they can operate.

The programmer makes room

Douetteau can be disarming about his own relationship with code. He has described loving Python while acknowledging that his coding skills have deteriorated. There is a small joke in the technical founder who now understands, personally, why someone might prefer an easier interface. A career can make an advocate of the user it once took for granted.

The same reluctance to play the all-purpose expert appears in his account of leadership. Asked about learning the commercial side of running Dataiku, he stresses listening, finding people with relevant experience and building a team he can rely on. Technical competence does not automatically supply every other competence a growing company requires.

When asked whom he looks to for inspiration, he declines to nominate a single role model. He reads fiction and nonfiction. “Inspiration comes from good stories, I think.” It is a pleasing answer from someone working in a field so fond of prediction. Stories admit motives, misunderstandings and competing points of view. Companies tend to contain all three.

“Inspiration comes from good stories, I think.”

Florian Douetteau

Growth, with a date attached

On October 17, 2025, Dataiku announced that it had passed $350 million in annual recurring revenue. It reported more than 750 customer organizations and more than 1,250 employees across thirteen offices and remote locations. Those are company-reported figures from that date, a snapshot of how far the original collaboration problem had travelled.

The company’s 2024 impact report also placed responsible and inclusive uses of AI among its stated commitments. Such promises become more consequential as software reaches more people. A founder can explain a philosophy in an interview; a company has to give it some expression in what employees build, what customers can inspect and how decisions get made. That gap between an intention and a working practice is familiar territory in Douetteau’s account of enterprise software.

The figures describe different obligations. Revenue represents a continuing commercial relationship. Customers bring different systems and expectations. Employees introduce their own coordination problem. The founder who wanted technical and business teams to understand each other must now run an organization in which that understanding has to survive distance and growth.

In June 2026, Maxwell Long joined as president and chief revenue officer, reporting to Douetteau. Long brought experience from Microsoft, Adobe and NetApp. The appointment gives a concrete expression to Douetteau’s preference for surrounding himself with people who know their jobs. Delegation sounds modest until the organization becomes large enough to make it indispensable.

The agent gets a performance review

By September 2026, the management question had acquired a new employee: the AI agent. Dataiku announced Agent Management, a standalone product intended to discover agents across different platforms, measure their performance and flag risk. The September announcement scheduled general availability for October. Counting the machines had become a business problem in its own right.

That month, a Dataiku and Harris Poll study of 685 CIOs reported that 81 percent lacked complete oversight of agents built outside approved systems or formal channels. The scope of that finding matters. It concerns a particular visibility gap in a surveyed group, rather than a census of every company’s AI. Even with that qualification, it makes an awkward companion to promises of effortless autonomy.

Douetteau put the management dilemma neatly: “almost nobody can fire an agent.” The joke lands because it borrows an old office ritual for a new kind of software. A company ought to know who owns a system, what it does and when to stop it. Giving software more freedom makes those questions harder to postpone.

At the same time, Dataiku expanded Cobuild, its natural-language interface, to help people move from business questions into governed data and AI work. Accessibility and oversight were developing together. The convenience of asking a question in ordinary language still needed something inspectable behind the answer.

What the business already knows

In a November 2025 conversation with Nicolas Marchais, Douetteau returned to the knowledge held by people inside an organization. Understanding the actual process, including its exceptions and unwritten judgments, is central to changing it. Much of what a business knows has never been neatly entered into a dataset.

His aspiration reaches beyond making existing tasks faster. He describes a future in which companies explicitly manage how they reason and make decisions. An organization’s particular way of doing business would become something it can express, examine and maintain as AI takes on more work. That is a demanding ambition for software, because an organization’s habits rarely arrive with a tidy instruction manual.

There is a tension here that makes his story interesting. A platform promises a common environment, while each business brings its own habits and demands. Douetteau’s approach asks for both: a shared way to build and oversee AI, with room for the particular knowledge that makes one organization different from another. Success depends on how well those two ambitions coexist when the system is doing real work.

His impatience with loose language is visible too. In a LinkedIn reflection after HumanX, he questioned the usefulness of the word agent, comparing its breadth with the early word website. What matters, he argued, is the specific form that follows the general label. The terminology will change; the task still needs to be understood.

Back at the beginning of this story sits the proposed application with the expensive running cost. It is a small incident beside years of company building, but it catches something distinctive about Douetteau’s position. He can imagine a much larger role for AI while asking it to account for itself in the present. The future may be intelligent. It will still have to get its expenses approved.