Dataiku has launched a separate product, Agent Management, to track and manage enterprise AI agents scattered across multiple platforms in one place, SiliconANGLE reported on Sept. 24 local time.

The company said large companies often manage software asset owners and license renewal timelines, but frequently do not know who operates AI agents and what risks they carry.

In an IBM survey cited by Dataiku, fewer than 1 in 5 organisations maintained a complete and up-to-date inventory of AI systems.

Agent Management connects to multiple execution environments to build a single inventory of agents. It can cover agents built on Salesforce Agentforce and services from Amazon Web Services, Microsoft and Google, as well as agents built on Databricks and Snowflake. Dataiku’s own agents and custom environments can also be connected via the OpenTelemetry standard.

Each agent entry displays the tools and models it depends on. Administrators can use that to check how it works. For higher-risk agents, authentication status and identified risks are recorded. Users can also ask questions in natural language, such as where risks are concentrated and which agents deliver low performance relative to cost.

Dataiku also expanded CoBuild, a natural-language tool for building AI projects that it officially launched in June. A newly added CoBuild Insights feature helps general employees ask questions of company data under governance and receive analyst-level answers. Dataiku Headless enables CoBuild to be used inside coding agents such as Anthropic Claude Code, OpenAI Codex and Cursor.

Developers can create or modify pipelines, models and agents in the CoBuild environment. It can also be used to convert legacy Microsoft Excel workbooks or Alteryx workflows into governed projects, or to investigate issues in a live environment. CoBuild is also embedded in Agent Management to support recommendations for business value metrics and diagnosis and remediation of risk alerts.