Dataiku Wants to Count the Agents CIOs Lost Track Of

Enterprises spent the last year building AI agents on every platform they already paid for. Now many of them cannot say how many they have. On September 24, at its annual Succeed conference, Dataiku launched Agent Management, a standalone product that finds every agent a company is running, whatever platform built it, measures how each one performs, and flags the ones that carry the most risk. It becomes generally available in October.

The same morning, Dataiku released a Harris Poll survey of 685 CIOs. Its headline number: 81% said they have lost oversight of their own AI agents. Dataiku CEO Florian Douetteau summed up how that happened in the launch release: "Nobody set out to build it this way. Teams built agents faster than anyone could count them."

What the product does

Agent Management connects to nine environments, per the Business Wire release: AWS Bedrock, Databricks Agents, Google Vertex, Microsoft Copilot Studio, Azure Foundry, Salesforce Agentforce, Snowflake Cortex, Dataiku's own platform, and custom environments through OpenTelemetry. Customers do not need to run the Dataiku platform to use it.

Once connected, it builds an inventory of agents and automatically identifies each one's structure, including the tools and models it uses. It tracks business and technical performance, flags the highest-risk agents, and keeps a standing record of certification status and named risks. Agents that handle customers, sensitive data or live transactions get priority. Recurring compliance tests run against certified agents, and administrators can query the whole portfolio in plain language.

Pricing is an annual per-instance fee plus metered monitoring charged per agent.

The distinction Douetteau draws is between watching agents and governing them. "Monitoring tells you an agent is running," he said in the survey release. "Managing tells you whether it's earned the right to keep running." Observability tools already exist on most of these platforms. Dataiku is selling a decision layer on top: which agents should be kept, fixed or retired.

What the survey says

The Harris Poll surveyed 685 CIOs online between July 9 and 29, 2026, across the United States, United Kingdom, France, Germany, the UAE, Japan, South Korea and Singapore. The survey was released by Dataiku to support a product launch, which is worth keeping in mind. The figures are still revealing.

Per the release, 84% of CIOs agree employees are creating agents faster than IT can govern them. 83% lack standardized lifecycle management for agents. 67% estimate they have more than 51 agents in production. 47% have already decommissioned more than 20 agents this year. Only 21% say they have full visibility into their AI costs, and 60% lack a central governance layer.

One gap stands out. 90% of CIOs say they are confident they track all their agents, yet 81% say they have lost oversight of them, and 72% cannot consistently measure the business outcomes agents produce. Those answers are hard to reconcile. Our reading is that most CIOs can list their agents but cannot say what those agents do or whether they are worth their cost. Only 28% measure both operational performance and business outcomes.

The pressure is personal. 76% said their role is at risk if AI does not produce measurable gains by the end of 2027, and 72% expect their AI budget to be cut or frozen if targets are missed by the end of this year. A CIO facing that deadline without a reliable way to show which agents produce value is exactly the buyer Dataiku is targeting.

The launch release also cites IBM research finding that fewer than one in five organizations keep a complete, current inventory of their AI systems.

Cross-platform against native

Every large platform on Dataiku's integration list sells its own governance. Salesforce has the Einstein Trust Layer, Microsoft has Agent 365 and Purview, and AWS builds controls into Bedrock. As Forkast notes in its analysis of the launch, the problem is that each one covers only its own agents. A company running agents on Agentforce, Copilot Studio and Bedrock ends up with three governance views and no single one.

Dataiku's bet is that multi-platform is the norm and will stay that way. The survey supports that: 75% of CIOs plan to use more or different models, 74% are considering open-source or open-weight models, and 91% favor letting teams build in a distributed way over central control. If building stays spread out, oversight has to sit above the platforms.

Dataiku is not alone there. Forkast lists Okta, IBM watsonx Orchestrate, Broadcom and SAP among vendors pursuing cross-platform agent control. The analysis also raises the obvious objection: buying another vendor to manage your vendors adds another layer of sprawl. It notes too that general availability moved from September to October, a small slip for a product built on the promise of reliability.

Why it matters

The agent market has moved from how to build agents to how to account for them. Two days earlier, Proofpoint announced a system of three autonomous agents built to secure data and AI activity as one risk, citing its own research that 87% of organizations have moved AI assistants beyond pilot. Agents are becoming a line item with owners, budgets and failure rates, and the 47% of CIOs who retired more than 20 agents this year show how quickly that line item changes.

The governance products that win will be the ones that answer the question boards ask: which agents are worth what they cost. Dataiku is early to frame the pitch around that question. Whether a data-science platform company can become the neutral registry for agents built on its competitors' clouds is the open test, and October's general availability is where it starts.