Dataiku's AI Agent Inventory Wants to Count Every Enterprise Agent

Enterprises are running AI agents they can't count

Enterprises cannot produce a reliable AI agent inventory — and the number of agents escaping oversight is growing fast. On September 24, 2026, at its Succeed conference in New York, Dataiku launched Agent Management, a standalone product built to find every AI agent an enterprise is running, measure how they perform, and flag the ones carrying the most risk. The product is scheduled for general availability in October 2026.

The core problem is simple to state and hard to solve: AI agents are being built and deployed far faster than anyone can catalog them. Dataiku's co-founder and CEO Florian Douetteau put it bluntly in the launch announcement: "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 or a guess."

That shrug is backed by research. IBM's "AI in Motion" study found that fewer than one in five organizations maintain a complete, current inventory of their AI systems — a statistic Dataiku cited directly when announcing the product, according to Pondero.

What the AI agent inventory actually does

Agent Management connects to the platforms enterprise teams already use to run agents: AWS Bedrock, Databricks Agents, Google Vertex, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, and Snowflake Cortex, plus Dataiku's own platform. For custom environments, it adds OpenTelemetry support. That connector list matters because it is where the cross-platform blind spot lives.

Once connected, the system scans each source into a single inventory. Every agent record captures the owner, purpose, connected systems, and last run, Aree Blog reports. More importantly, Agent Management automatically identifies each agent's structure — including the tools and models it relies on — so supervisors see how an agent actually works, not just that it exists, according to Help Net Security.

The product is designed as more than a directory. It monitors each agent's health, usage, cost, quality, and behavior over time, with alerts when an agent drifts away from its normal pattern. Agents that touch customers, sensitive data, or live transactions get a standing record of certification status, named risks, and scheduled tests. Dataiku identified examples of risks worth naming: excessive privileges, shared credentials, and the absence of a human override. The intent is that an evidence trail already exists when a manager, auditor, or regulator asks for it.

Why native dashboards can't see the whole estate

Most enterprise agent governance today lives inside the platform that built the agent. Salesforce's Einstein Trust Layer covers Agentforce agents. Microsoft's Agent 365 and Purview cover Copilot agents. Neither can see past its own walls, Forkast reported. Dataiku is betting that a product sitting above all of those platforms at once has a market, precisely because none of the platform vendors can build a tool that also inventories a competitor's agents.

The scale behind that bet is striking. Forkast, citing Gartner, reported that the average Fortune 500 company will manage more than 150,000 AI agents by 2028, up from fewer than 15 in 2025. An OutSystems survey of 1,900 IT leaders found that 96 percent of enterprises already run agents while only about 12 percent have a centralized way to manage them.

That gap between deployment and oversight is not evenly distributed. A team running agents on three or more of the major platforms has exactly the cross-platform blind spot Dataiku is describing. A shop running agents only inside Salesforce or only inside Microsoft Copilot Studio is arguably already getting equivalent oversight from that platform's own trust layer, Forkast's analysis noted.

What to watch before October

Two things will decide whether this becomes a standard part of the enterprise stack. The first is price: Dataiku has not published a per-instance figure, and pricing is annual per instance with monitoring metered per agent. Anyone budgeting for agent governance should wait for that October number before comparing Agent Management's cost against staying inside the built-in tools, LapaaS Voice noted.

The second is whether the cross-platform inventory works as described outside a conference stage. General availability in October is when the first customers beyond Dataiku's own launch event will confirm that. The broader trend is clearly moving toward tighter accountability — a theme that also runs through my recent coverage of OpenAI's rogue agents probing government sites and the weekend fallout from that episode, as well as the intensifying consumer AI agent race between ByteDance's Doubao and Meta's Muse. As agents take on more consequential work, knowing where they are and what they're doing is becoming the baseline expectation rather than a luxury.

Dataiku's bet is that the inventory problem comes first — before quality testing, cost comparison, or deciding what to retire. For enterprises where the answer to "how many agents do you run?" is still a shrug, that may be exactly right.

Sources: Aree Blog; Pondero, citing Dataiku's September 24, 2026 announcement, Forkast, Gartner, OutSystems, and IBM Institute for Business Value; Help Net Security; BigDATAwire; LapaaS Voice.