Dataiku launches Agent Management to rein in enterprise AI agent sprawl

Dataiku has launched Agent Management, a standalone product designed to find every AI agent running inside a large organisation, measure what each one is worth, and flag the ones that pose the biggest risk. The Paris-founded enterprise AI vendor unveiled the product on 24 September 2026, with general availability following in October.

The launch addresses a problem that has crept up on large enterprises largely unnoticed: nobody can say with confidence how many AI agents they actually have running, who built them, or what they are doing with company data. Agent Management is Dataiku's attempt to become the inventory and oversight layer that answers that question, regardless of which vendor's tools were used to build the agents in the first place.

Why counting agents got hard

Florian Douetteau, Dataiku's co-founder and chief executive, framed the problem in blunt terms. "Ask a bank how many servers it runs, and you get an answer to the decimal. Ask how many AI agents it's running, and you get a shrug or a guess," he said. He added that the gap was not deliberate: "Nobody set out to build it this way. Teams built agents faster than anyone could count them. Agent Management tells you what's actually out there, and what it's actually worth."

That scattergun growth is the result of different teams inside the same company building agents on whatever platform was closest to hand. A sales team might stand up an agent in Salesforce Agentforce, a data science group might build one in Databricks, and an IT department might lean on Microsoft Copilot Studio or Azure Foundry. Each tool reports on itself well. None of them, on its own, can tell a chief risk officer how many agents the whole company is running or which ones touch sensitive data.

What the product actually does

The product plugs into the major clouds and vendor tools where companies actually build agents today: Amazon's Bedrock, Databricks, Google's Vertex AI, Microsoft's Copilot Studio and Azure Foundry, Salesforce's Agentforce, Snowflake's Cortex, and Dataiku's own stack. It also speaks OpenTelemetry, a widely used open monitoring standard, so it can pick up custom, home-built agents that were never built on a named vendor platform at all.

Once connected, the product scans all of those sources into a single inventory and automatically maps the structure of each agent, including which underlying models and tools it calls on. That matters because knowing an agent exists is different from knowing how it actually works, and supervisors need the latter to judge whether it is safe to leave running.

For agents judged highest-risk, meaning those handling customers, sensitive data or live transactions, Agent Management keeps a standing record of certification status and named risks, and reruns tests on a schedule. The aim is to have an evidence trail already assembled before a manager, internal auditor or outside regulator asks for one, rather than scrambling to produce it after the fact.

Dataiku describes the product as deliberately vendor-agnostic. Sitting above individual platforms lets it answer questions that no single vendor's dashboard can, such as which agents nobody is actively monitoring, where risk is concentrated across the business, and which agents are actually earning back what they cost to run. Teams can put those questions in plain language and get a portfolio-wide answer rather than checking five separate tools.

Part of a bigger governance push

Agent Management is not Dataiku's first product aimed at the agent boom this year. In June 2026 the company released Cobuild, a tool that turns plain-language business objectives into AI projects without requiring code, while still routing the result through approval workflows rather than letting it bypass internal controls. In March, Dataiku rebranded its core platform around what it called the Platform for AI Success, adding "Reasoning Systems" aimed at manufacturing, supply chain and financial-risk use cases.

Taken together, the three launches point to a company betting that the next phase of enterprise AI spending will go toward oversight and control rather than toward building yet more agents. For European banks, insurers and industrial firms already under pressure from the EU AI Act to document and justify their automated systems, a tool that produces a ready-made inventory and audit trail has an obvious appeal.

Dataiku has priced Agent Management per instance annually, with monitoring metered per agent on top, a structure that will let larger organisations scale cost with how many agents they actually track. Whether enterprises treat agent governance as a line item worth buying separately, rather than a feature bundled into the platforms they already pay for, is the open question the October rollout will start to answer.