Dataiku announced Agent Management on September 24 at its annual Dataiku Succeed conference, a standalone product built to find every AI agent an enterprise is running regardless of which platform built it, measure each one’s business and technical performance, and flag the agents that carry the greatest risk. The product connects to eight major agent-building platforms, AWS Bedrock, Databricks Agents, Google Vertex, Microsoft Copilot Studio, Azure Foundry, Salesforce Agentforce, Snowflake Cortex, and Dataiku’s own platform, plus OpenTelemetry support for custom environments, and does not require an enterprise to run its agents on Dataiku itself. General availability is set for October 2026.
The problem Dataiku says it is solving
CEO Florian Douetteau put the gap 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.” Dataiku cites IBM research finding that fewer than one in five organizations maintain a complete, current inventory of the AI systems they operate. The structural reason, per Dataiku, is that individual agent-building platforms only have visibility into the agents built on that platform; a large enterprise running agents across Salesforce, Microsoft, AWS, and several other tools has no single system that sees all of them together, which means no single system can price their risk or their value across the business either.
The product itself is built around that gap specifically. Rather than requiring an enterprise to route its agents through Dataiku’s own platform to get visibility, Agent Management reads directly from the platforms enterprises already use, plus OpenTelemetry for anything custom. That design choice is itself a statement about where Dataiku sees the market heading: multi-platform agent sprawl is the default condition for any enterprise past a certain size, not a temporary state that consolidates around one vendor, and a governance product that assumes otherwise would already be built for a market that no longer exists.
Why it matters to the enterprise AI buying committee
A cross-platform agent inventory tool is itself evidence of where the market has moved: agent sprawl, not agent capability, is now the operational problem enterprises are paying to solve. Three implications worth weighing:
- “We don’t know how many agents we’re running” is now a fixable gap, not an inherent limitation. If a purpose-built inventory product can connect to eight major platforms and produce an answer, an enterprise citing platform fragmentation as the reason it lacks agent visibility no longer has that excuse available to its board or risk committee.
- Per-agent ownership and cost data changes who gets asked to justify an agent’s existence. Dataiku’s product records each agent’s owner, purpose, usage, cost, and quality; that level of accountability shifts agent governance from an abstract policy exercise to a line-item conversation with whichever business unit owns each agent, similar to how cloud cost allocation eventually forced accountability onto individual engineering teams.
- This is a genuinely new product category, not a feature bolted onto an existing platform. Dataiku is selling agent inventory and risk tiering as a standalone product independent of its own core platform, a bet that enterprises will pay separately for cross-platform oversight rather than expect it bundled into whichever agent-building tool they already use most.
The read this announcement doesn’t spell out
Dataiku’s own framing, that fewer than one in five organizations can answer a basic inventory question about their AI systems, lands the same month Microsoft restructured its own Responsible AI Standard specifically around agentic-system risk and roughly six weeks after Deloitte’s own survey data showed a wide gap between enterprises that have tested agents and those that have scaled them under real governance. Taken together, the pattern across all three pieces of evidence points at the same conclusion from different directions: agent governance tooling, not agent capability, is now the binding constraint on how far enterprises can actually scale agentic AI.
Source: Dataiku Newsroom
