Introduction
Ask any Chief Data Officer how much data flows through their pipelines and you'll get a precise answer. Ask how many AI agents are actively running across their organization, and things get vague fast. According to a 2026 study, fewer than one in five companies have a complete inventory of their AI systems. That's the gap Dataiku set out to close with the launch of Agent Management on September 24, 2026.
Data Processes Without Lineage or Ownership
Data teams have spent years building catalogs, lineage tools, and data quality policies. They know which table feeds which dashboard, which pipeline transforms sensitive data, and who owns which dataset. That rigor — hard-won through regulation and incident response — breaks down the moment an AI agent enters the picture.
Yet an AI agent is a full-fledged data consumer. It queries databases, calls APIs, accesses CRM and ERP systems, and produces outputs that feed downstream decisions. But unlike a Spark job or a scheduled SQL query, its data access is invisible to traditional governance tooling. It operates outside the catalog, outside lineage tracking, often without a designated owner.
A 2026 survey of nearly two thousand IT leaders puts the scale of the problem in sharp relief: 96% of companies report running at least one AI agent in production, yet only 12% have a centralized solution to oversee them. The rest manage on a case-by-case basis — platform by platform, team by team.
A Multi-Platform Inventory as the Starting Point
Agent Management positions itself as a cross-cutting layer, independent of whichever platform was used to build the agents. The product connects natively to AWS Bedrock, Databricks, Google Vertex, Microsoft Copilot Studio, Azure Foundry, Salesforce Agentforce, and Snowflake Cortex. Custom environments can plug in via OpenTelemetry.
Onboarding is deliberately straightforward: select the platform, authenticate, and the system automatically scans for active agents. It then surfaces their internal structure — tools used, underlying models, dependencies — while tracking certification statuses and scheduling continuous testing.
What sets this approach apart from conventional monitoring tools is the business dimension. Agent Management doesn't just check whether an agent is responding; it measures whether the agent is delivering expected outcomes, bridging technical metrics with business KPIs. Few data observability tools offer that level of integration today.
What This Means for Data Teams
For governance leads, the stakes are real. AI agents represent a new class of data asset — on a par with tables, ML models, and dashboards — but without the registration, validation, and oversight processes that govern those assets. No inventory means no lineage: there is no way to trace which sensitive data an agent accessed or which decisions it influenced.
General availability for Agent Management is set for October 2026, with per-instance annual pricing and consumption measured per agent. The product is standalone — no existing Dataiku platform installation required.
For CIOs and Chief Data Officers, the question is no longer whether their AI agents deserve governance. It's figuring out how many of them are already running without anyone knowing.