Dataiku Agent Management Aims to Inventory Every Enterprise AI Agent

Enterprises have spent the last two years building AI agents at breakneck speed. Now a new question is landing on chief information officers' desks: how many agents do they actually run? Dataiku, the AI platform company, thinks the answer is usually a guess — and its answer to that problem is Dataiku Agent Management, a new cross-platform agent governance product announced on September 24, 2026, at the company's Succeed conference in New York.

The product is a standalone tool that finds, measures, and flags risk in AI agents regardless of which platform built them. General availability is set for October 2026, which puts it on shelves just as many enterprises move from agent pilots to production fleets.

The inventory gap is getting dangerous

The numbers behind the launch are stark. Fewer than one in five organizations maintain a complete, current inventory of their AI systems, according to research from the IBM Institute for Business Value that Dataiku cited in its announcement.

That blind spot compounds as agent counts explode. Gartner, as reported by Forkast, projects 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 adds texture to the gap: 96 percent of enterprises said they already run AI agents, while only about 12 percent said they have a centralized way to manage them. The mismatch is where risk lives — agents that touch customers, move money, or read sensitive data without any named owner or standing certification record.

Dataiku chief executive Florian Douetteau framed the problem as an inventory failure rather than a technology failure. A bank can count its servers down to the decimal, he said, but asking the same bank how many AI agents it runs produces only a shrug or a guess. Teams built agents faster than anyone could count them, and governance has not caught up. The timing fits a wider regulatory mood: earlier coverage on this site examined how regulators are beginning to probe the risks of unmanaged AI systems (FTC Rogue AI Probe Targets OpenAI and Anthropic).

What Dataiku Agent Management actually does

Agent Management connects to six of the major enterprise agent platforms: AWS Bedrock, Databricks Agents, Google Vertex AI, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, and Snowflake Cortex. It also supports Dataiku's own platform and adds OpenTelemetry support for custom environments.

The product builds a single inventory of every connected agent and maps the tools and models each one relies on. For agents that touch customers, sensitive data, or live transactions, it keeps a standing record of certification status, named risks, and scheduled tests.

The idea is that when a manager, auditor, or regulator asks what a given agent is allowed to do, the audit trail already exists rather than having to be reconstructed after the fact. The design deliberately measures both business and technical performance, not just uptime or latency.

Dataiku's pitch is that an inventory tells an enterprise what is actually out there and what it is actually worth — which agents justify their cost, which carry unacknowledged risk, and which duplicate each other across platforms.

The bet: cross-platform visibility beats native tools

The launch is also a strategic wager. Most agent governance today lives inside the platform that built the agent. Salesforce's trust tooling covers Agentforce agents, and Microsoft's compliance stack covers Copilot agents, but neither can see past its own walls.

Dataiku is betting that a neutral product sitting above all of those platforms has a market, because no platform vendor can build a tool that also inventories a competitor's agents. That logic has limits, and buyers should apply it carefully.

A company running agents on three or more of the connected platforms has exactly the cross-platform blind spot Dataiku is describing. But an organization whose agents live entirely inside Salesforce or entirely inside Microsoft's stack is already getting equivalent oversight from those platforms' own trust layers — and may not need a second governance product at all.

Pricing will decide how quickly the bet pays off. Dataiku says the product is priced per instance annually with monitoring metered per agent, but it has not published a specific dollar figure. Anyone budgeting for agent governance should wait for the October general-availability pricing before comparing the cost against staying inside their platform vendors' built-in tools.

Why this launch matters for the agent economy

Dataiku Agent Management is part of a broader shift: agent governance is becoming a product category of its own, separate from the platforms that build the agents. Fresh capital keeps flowing into the space, with agentic security firms raising new rounds and enterprises waking up to the risks of unmanaged agent fleets.

The governance question is arriving just as the agent stack gets real hardware of its own, with chips and on-device runtimes built to run agents locally rather than in the cloud (Nvidia N1X PC Chip Sinks Intel and AMD Stocks). Tools that count, certify, and monitor agents are the enterprise's answer to a world in which software writes and runs itself.

The date to watch is the general-availability launch itself, when the first customers outside Dataiku's own conference stage will confirm whether the cross-platform inventory works as described. If it does, October 2026 may be remembered as the month the agent inventory went from a shrug to a system of record.

Sources: Dataiku press release via Business Wire (September 24, 2026); reporting and analysis via Pondero; IBM Institute for Business Value research on AI systems inventory; Gartner agent-count projections and OutSystems survey data as reported by Forkast.