Product1 publisher3 min readPublished
Dataiku sells an agent inventory to the nine in 10 CIOs who say they already have one
Dataiku ships Agent Management in October with connectors into Salesforce, AWS, Microsoft and Google agent runtimes, metered per agent monitored. The bill grows with every agent the inventory turns up.
The Product Desk · Product desk
What happened
- Dataiku launched Agent Management, a standalone product that finds and tracks every AI agent a business runs, including agents built on other vendors' platforms, with general availability due in October.
- Access will carry an annual per-instance fee with monitoring metered by the agent, and Dataiku gave timing and pricing for Agent Management only.
- Nine in 10 of the 685 CIOs polled for Dataiku by The Harris Poll said they were confident they had complete tracking of their agents.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- contradiction Dataiku is selling into a gap its own survey says buyers do not have, and the 70-point distance between 90% CIO confidence and IBM's sub-20% inventory figure comes from different populations, so a buyer is left guessing which number describes their own estate.
- cost The budget line can only be sized once the first inventory has run, because the meter counts whatever that scan turns up.
- constraint Completeness stops at the connector sheet, so the unapproved agents a CIO is least sure about only surface after someone instruments the environment they run in.
- decision Teams already watching agents in each cloud's native console now have to decide whether a layer above all of them is worth a second contract.
A bank can say how many servers it runs "to the decimal," according to Dataiku co-founder and chief executive Florian Douetteau, while the same bank asked about its AI agents gives "a shrug or a guess" [8]. Dataiku's own survey says something close to the reverse. Nine in 10 of the 685 CIOs The Harris Poll surveyed for it said they were confident they had complete tracking of their agents [10]. In IBM research the company cites, fewer than one in five organizations keep a complete and current inventory of AI systems [7]. The gap is more than 70 percentage points, and the two figures count different populations in different surveys [17]. Dataiku's report also asked CIOs about agents built outside approved systems or formal channels; Dataiku did not publish that answer [18].
Discovery is conditional. Once connected to the platforms where a company's agents run, the software pulls them into a single inventory [5]. Connectors cover Salesforce's Agentforce and the agent services from AWS, Microsoft and Google, agents built on Databricks and Snowflake show up alongside Dataiku's own, and custom environments report in through OpenTelemetry [4]. An agent running somewhere nobody has wired up does not appear, and wiring up a custom environment is work a customer's engineers do first.
Access carries an annual per-instance fee with monitoring metered by the agent [3]. The structure ties Dataiku's revenue to the number CIOs say they already have: every agent it surfaces on a competitor's runtime adds to the meter [19].
What each entry holds is more useful than the count. The inventory maps the tools and models an agent depends on, the riskiest agents carry a standing record of certification status and named risks, and their tests rerun on a schedule so an evidence trail exists before an auditor asks for one [6]. The person served here is whoever has to produce that file. Dataiku pitches the product against monitoring built into a single vendor's stack, which it argues favours that vendor's own agents [11].
The Cobuild half is aimed at different people. Dataiku Headless puts Cobuild inside Claude Code, Codex and Cursor, where a developer can build or change pipelines, models and agents [13], while Cobuild Insights lets any employee query governed company data for what Dataiku calls "analyst-quality answers" [14]. An old Excel workbook or Alteryx workflow can be converted into a governed project [15]. Dataiku gave timing and pricing for Agent Management only [2].
Clement Stenac, Dataiku's co-founder and chief technology officer, said enterprises tend to run technical and business teams on separate platforms "without shared data or shared guardrails," which a few pilots can live with but at scale becomes "a liability companies have to prepare for" [16].
Every runtime where an agent could be running divides into two groups: the ones on the connector sheet and the rest. The first group reports in when the contract starts. The second is where the agents a CIO was guessing about live, and it is also where a customer's own engineers have to go before any inventory, from any vendor, can count them.
What to watch
- Whether Dataiku publishes the survey's answer on agents built outside approved systems or formal channels. That is the number its pitch rests on.
- Whether the connector list grows by the October general availability date, and whether Salesforce, AWS, Microsoft or Google add cross-vendor agent inventory to their own consoles.
- Published figures for the annual per-instance fee and the per-agent monitoring rate. With those, a buyer could model the bill before scanning.