Dataiku sells agent inventory that companies have unknowingly deployed on eight competing platforms

The product scans Bedrock, Databricks, Vertex, Copilot Studio, Azure Foundry, Agentforce, Snowflake Cortex, and OpenTelemetry-instrumented environments, and then classifies each discovered agent by risk level.

Announced on September 24, 2026 and expected to be generally available in October, Agent Management is sold by Dataiku as a standalone product, independent of its own platform. Its principle: to browse a company's cloud and SaaS environments to identify the AI agents that run there, regardless of the tool with which they were built. The list of connectors covers AWS Bedrock, Databricks Agents, Google Vertex, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, Snowflake Cortex, Dataiku itself, and OpenTelemetry for in-house development.

The product doesn't just count. For each agent discovered, it automatically reconstructs the structure — the tools mobilized and the models called — so that a manager can see how the thing really works, not just see that it exists. The agents are then broken down by level of risk, and for the most sensitive ones — those that affect customers, personal data, or transactions in production — the tool keeps a permanent record: certification status, identified risks by name, and replayed tests according to a schedule, so that the audit trail already exists on the day an auditor or regulator requests it.

The need for this type of tool is measurable: according to IBM's "AI in Motion" study, fewer than one in five organizations maintain a complete and up-to-date inventory of their AI systems. In a company that has let multiple teams build agents on three or four different platforms, the most mundane question — who owns that agent, whether it achieves its goals, whether it needs to be fixed — often has nowhere to ask.