At the **Dataiku Summit** in Paris, the company unveiled two major updates to its platform: the **Agent Management** solution and an evolved interface called **Cobuild**. These were presented as key components of the strategic vision outlined by Florian Douetteau, co-founder and CEO of Dataiku, on September 29. Douetteau emphasized that the success of artificial intelligence (AI) now hinges on process transparency and control over the infrastructure that supports it. He highlighted the need to balance innovation with the stability of existing operations, particularly as AI deployment in enterprises has matured. While AI is generating value, it is unevenly distributed, with about 20% of companies capturing 75% of the benefits, according to Douetteau. Douetteau attributed this imbalance to a common mistake, which he called the "supermarket model." This refers to the tendency of companies to buy pre-made AI models, treating them as off-the-shelf solutions to business problems. However, he explained, 80% of companies fail to achieve a return on investment (ROI) due to the difficulty of integrating these models into existing workflows. These models are often opaque, or "black boxes," and rely on technologies that may be unreliable or expensive to access. To address these issues, Dataiku proposed two contrasting strategies. "Strategy A" calls for radical technological change, outsourcing expertise, and replacing human processes with autonomous systems. Governance, in this case, is secondary. In contrast, "Strategy B," which Dataiku supports, focuses on developing internal skills and bridging the gap between traditional and new technologies. Under this approach, business experts—referred to as "guarantors of truth"—make the final decisions. This strategy requires a system that can track and justify every decision made by AI models, ensuring infrastructure independence. Dataiku's platform supports both public cloud services and private, sovereign infrastructures like **Scaleway** or **Orange**, which are crucial for industries with strict regulatory requirements. The company also highlighted the need for control tools, especially for managing AI agents. Douetteau warned that the uncontrolled use of AI within departments can lead to "shadow AI," where systems operate outside of oversight, risking the coherence of an organization's information systems. To combat this, Dataiku introduced **Agent Management**, set to launch on October 15, as a centralized tool to track and govern AI agents across different platforms. The system allows IT and business teams to create a structured and auditable portfolio of AI tools. It includes a centralized inventory that can register AI agents on various platforms through direct integration. Beyond basic monitoring, the service tracks detailed performance metrics. In a demo using a fictional company, the system detected that an AI agent exceeded its operational cost threshold by 10%. Further analysis revealed that 33 of the agent's requests were outside its intended scope, as it was used for tasks it wasn't designed for. Despite these deviations, the agent still delivered a measurable benefit: avoiding 20 hours of production downtime and saving nearly $19,000 over 30 days. Sophie Dionnet, SVP of Product and Business Solutions at Dataiku, explained that AI agents are too versatile to be treated like traditional IT assets. Their management requires governance policies to be translated into clear, traceable practices. The **Agent Management** tool aims to move from passive observation to active control through technical safeguards and the creation of corrective action tickets. This centralized approach, Dionnet said, is essential for estimating performance coherently and aligning AI innovation with financial goals. It also aims to simplify the process of creating new AI agents. The second major development from Dataiku is the **Cobuild** interface, which allows business users to create AI tools through conversational interfaces. The goal is to make it easier for non-experts to design complex data solutions. Unlike other conversational AI systems, **Cobuild** visually maps out each step of the data transformation process, creating a "flow" that users can inspect, audit, and modify. This "Human-in-the-loop" mechanism ensures transparency. IT teams can set secure limits in advance, and no AI asset created through Dataiku's tools can be deployed without approval. Dataiku claims that **Cobuild** currently has 2,000 active users across 250 accounts and aims to triple that number by year-end. The company argues that this approach will allow users to answer complex business questions in natural language, without needing to write SQL queries.