Ask a bank how many servers it runs and you get an exact number. Ask how many AI agents it runs and, in the words of Dataiku CEO Florian Douetteau, you get “a shrug or a guess.” That line comes from the September 24 launch announcement for Dataiku Agent Management, and it describes a problem many engineering leaders will recognize. The product matters less than what it implies: the AI agent inventory is turning into its own software category.
What Dataiku actually shipped
Agent Management is a standalone product; Forkast reports it works without Dataiku’s core platform. According to the company, it connects to AWS Bedrock, Databricks Agents, Google Vertex, Microsoft Copilot Studio, Azure Foundry, Salesforce Agentforce and Snowflake Cortex, with OpenTelemetry support for custom environments. General availability is planned for October 2026, priced per instance per year with metered monitoring per agent.
Per eWeek’s write-up, each inventory entry can include the tools and models an agent uses, its owner and purpose, usage and cost, plus risk assessments and certification status for higher-risk agents. Dataiku also cites IBM research finding that fewer than one in five organizations keep a complete, current inventory of their AI systems.
Why an inventory, and why now
Gartner’s April guidance on agent sprawl, as reported by Tech.co, projects that an average Fortune 500 company will run more than 150,000 agents by 2028, up from fewer than 15 in 2025. In the same coverage, only 13% of organizations say they have the right governance for agents today.
Those numbers explain the sequencing. Teams build agents inside whichever platform is closest: a Copilot Studio agent in one department, a Bedrock agent in another, a Salesforce agent owned by sales operations. Each platform sees its own agents. Nobody sees all of them. Gartner’s second step in its six-step framework is a centralized agent inventory, right after governance policy, which suggests you cannot enforce much until you know what exists.
What an inventory does not do
Read the limits before the pitch. eWeek reports that Dataiku’s tool is not an AI gateway sitting between agents and their tools, so it does not see real-time actions and cannot stop an agent mid-task. Dataiku’s own documentation, quoted there, says integration depth depends on what each connected platform exposes.
That is a real distinction. An inventory tells you which agents exist, who owns them and how risky they look. Runtime control, meaning permissions, rate limits and kill switches, sits elsewhere. Buying one and assuming you have the other is an easy mistake.
Cross-platform tool or native dashboards?
Forkast frames the launch as a bet that cross-platform governance beats native tooling. Vendors such as Salesforce and Microsoft offer deep controls inside their own ecosystems, but those controls leave gaps across a mixed estate. The counterpoint Forkast raises is fair: adding a vendor to watch your other vendors is its own overhead. The answer depends on how many platforms you really run. With two, native tools may be enough. With five, a spreadsheet stops working.
Start with a minimal record
You can begin before any purchase. For every agent, capture the following:
- A named human owner and the business purpose
- The platform it runs on and the models and tools it can call
- The data it can read and the actions it can take
- Risk tier, last review date and monthly cost
Gartner’s steps on identity, permissions and lifecycle build on exactly this record. If an agent has no owner, that is your first finding.
The takeaway
Servers got asset management and cloud accounts got cost tooling once counts got out of hand. Agents are following the same path, only faster. Whether you choose Dataiku, a native dashboard or a well-kept register, decide who owns the list before the count reaches four digits. If you are building agents and want help designing the architecture and controls around them, talk to the Luby team.