Artificial intelligence data analytics startup Dataiku Inc. is evolving its popular data science studio to focus more on the development of trusted and measurable autonomous AI agents.

The startup today unveiled a new Platform for AI Success that merges its existing data science platform for developing AI models with three entirely new products: Dataiku Agent Management, Dataiku Cobuild and Dataiku Reasoning Systems, which combine to transform the way AI agents are built, controlled and scaled.

Dataiku co-founder and Chief Executive Florian Douetteau told SiliconANGLE that AI’s adoption is spreading so fast that enterprises are struggling to keep control. Most organizations operate across fragmented, multi-vendor environments, resulting in their AI agents duplicating work with inconsistent performance and tons of governance blind spots, creating unacceptable operational risk. Worse, they’re doing this without any kind of control layer, which means they struggle to prove the impact of AI agents, manage their costs or defend decisions made by those systems.

“Without bringing everyone into the building process, AI initiatives won’t be relevant or accepted; without orchestrating complex, modern technologies, AI will be too naive to have a meaningful impact; and without governing AI at every single step, it will never move beyond the proof-of-concept phase,” Douetteau said. “We built our platform specifically to solve this exact roadblock.”

Dataiku’s Platform for AI Success intends to become that missing layer, connecting data platforms with enterprise systems, foundational models and third-party agent frameworks through a single governed environment. It’s not available yet, but it will be starting in April with the launch of a first core capability called Dataiku Agent Management, which is an agent monitoring tool that aims to help enterprises gauge the value of each AI agent they deploy.

It will be made available as a standalone product, giving enterprises the ability to “see” every agent it deploys across multiple information technology environments and measure its business impact. Each AI agent will be evaluated against defined key performance indicators, and its performance will be continuously monitored to alert organizations to problems such as model drift and cost overruns.

Douetteau said Dataiku Agent Management will enable organizations to gain full visibility over which AI systems they have running, what decisions those systems are making, and critically, provide them with the information they need to decide if it’s actually worth bothering with.

A second key capability will arrive later this year with the launch of Dataiku Reasoning Systems, which will help companies to transform their data, business rules and human-defined decision logic into an operational intelligence environment that feeds into their AI agents. So instead of deploying standalone agents to perform discrete tasks, enterprises will be able to orchestrate fully-governed “decision systems” that embed company- and industry-standard reasoning into their workflows.

Finally, Dataiku Cobuild will provide a visual interface that enables users to describe business objectives in natural language and automatically generate AI agents that can achieve them, complete with all of the required data pipelines, models and applications. For each new agent, Dataiku Cobuild will generate a structured visual flow that users can review step-by-step to validate assumptions and decision-making logic and approve before they’re deployed in production.

Launching in June, Cobuild will essentially provide a way to transform business intent into “executable logic.” Dataiku will handle all of the complex backend tasks, such as configuring environments, resources and controlled, repeatable deployment.

By giving organizations the ability to create and coordinate intelligent agents across any environment more easily, Dataiku says, it’s empowering every business expert to contribute to the agentic revolution.

Co-founder and Chief Technology Officer Clément Stenac said agentic AI urgently requires “structured orchestration,” so agents are fed the right data and controlled by a combination of business rules and human oversight. “That coordination layer is missing in most deployments, so the Platform for AI Success is designed to fill that void,” he added.

Image: SiliconANGLE/Meta AI