Dataiku has launched Agent Management, a standalone product designed to give enterprises a single inventory of the AI agents running across their businesses, regardless of where those agents were built.
The company announced the product on September 24, 2026, at Dataiku Succeed in New York City. Dataiku said Agent Management will be generally available in October 2026.
The product is designed to identify AI agents across multiple enterprise platforms, measure their technical and business performance, and flag agents that carry higher levels of risk. Dataiku’s launch announcement describes the product as a layer that sits above the platforms where agents are built and operated.
Dataiku said Agent Management connects to AWS Bedrock, Databricks Agents, Google Vertex, Microsoft Copilot Studio, Azure Foundry, Salesforce Agentforce, Snowflake Cortex and Dataiku. It also supports OpenTelemetry for custom environments.
Once connected, the system scans agents into a single inventory. Dataiku said each record includes information such as the agent’s owner, purpose, connected systems and last run. It can also identify the structure of an agent, including the tools and models it uses.
The company is positioning the inventory as more than a directory. Agent Management monitors an agent’s health, usage, cost, quality and behavior over time, with alerts when an agent moves away from its normal pattern.
For higher-risk agents, Dataiku said the product maintains records covering certification status, named risks and scheduled tests. The company identified examples of risks including excessive privileges, shared credentials and the absence of a human override.
Dataiku said those records are intended to provide a continuing history of certifications, tests and alerts for agents that handle customers, sensitive data or live transactions.
Agent Management does not operate or filter the agents themselves. Dataiku said the product observes, measures and records what the agents do while leaving their execution and decisions to the platforms on which they run. For agents orchestrated by Dataiku, runtime control remains with its LLM Mesh.
The product also allows users to ask portfolio-level questions in plain language, including which agents are not being monitored, where risk is concentrated and which agents are generating enough value to justify their cost.
Dataiku said Agent Management is priced through an annual fee per instance, with monitoring metered per agent. The company has not published numerical pricing in its launch announcement.
The launch follows Dataiku’s March 2026 introduction of its broader Platform for AI Success. Agent Management was announced then alongside Dataiku Cobuild and Dataiku Reasoning Systems as one of three new products.
Dataiku is launching the inventory as enterprises increase the number of agents they deploy across different teams and systems.
A September 24 study from Dataiku and Harris Poll found that 67% of surveyed CIOs estimated their organizations had at least 51 AI agents actively running in production. At the same time, 90% said they were confident they had complete tracking of those agents, while 72% said they could not consistently confirm whether all of them were delivering the business outcomes they were built to produce.
The same survey found that 60% of CIOs lacked a central AI governance layer covering enterprise applications, tools and IT environments, while 81% lacked complete oversight of agents created outside approved systems or formal channels.
Dataiku also reported that 84% of CIOs agreed employees were creating AI agents and applications faster than IT could govern them. Only 28% said they consistently measured both operational performance and business outcomes across all of their AI agents.
The survey found that 83% lacked standardized agent lifecycle management across their organizations, 47% had already decommissioned more than 20 agents during 2026, and only 21% reported full, near-real-time visibility into AI costs with attribution by business unit, team or use case.
CIOs also reported a move toward using multiple models. Seventy-five percent said they planned to use more or different models to improve AI continuity, 74% were considering open-source or open-weight models as protection against future model unavailability, and 80% had conducted or planned to conduct AI dependency risk assessments.
The study found that 91% of CIOs agreed the most effective AI strategy was not to centralize every application and agent, but to allow business teams to build within a governed environment.
The survey also found that 76% of CIOs said their organizations lacked mature, highly capable return-on-investment measurement across all or most AI initiatives. When agents fail, responsibility was divided among shared teams at 23%, central IT at 21%, data or AI teams at 20%, and security, risk and compliance at 18%.
The survey covered 685 CIOs in the United States, United Kingdom, France, Germany, the United Arab Emirates, Japan, South Korea and Singapore. Harris Poll conducted the online research for Dataiku from July 9 to July 29, 2026.
The report also found that 88% of CIOs said their professional reputation or career trajectory would be shaped by their success with AI, while 87% said their CEO had indicated their job security depended on AI outcomes. Seventy-six percent believed their own role would be at risk if their company failed to deliver measurable AI gains by the end of 2027.
Board pressure is also increasing. Dataiku reported that 97% of CIOs had seen at least some increase in board pressure to demonstrate measurable AI return, with 77% describing that increase as significant or moderate. Seventy-two percent said their AI budget was likely to be cut or frozen if performance targets were not met by the end of 2026.
Among U.S. CIOs, 94% said employees were creating AI agents and applications faster than IT could govern them, 88% believed their role would be at risk if their organizations failed to deliver measurable AI gains by the end of 2027, and 87% said their AI budget was likely to be frozen or cut if performance targets were missed. Ninety-two percent said they would stake their job on delivering measurable AI results, while 82% regretted at least one major AI vendor or platform selection made during the previous 18 months.
IBM research provides a wider measure of the inventory problem. IBM’s Institute for Business Value reported that only 18% of organizations maintain a current and complete AI inventory. Its 2026 technology-leader study also found that technology leaders expected to deploy an average of 1,661 AI agents by 2027, a 38% increase from the current level reported in the study.
IBM also reported that 85% of technology leaders lacked full visibility into real-time AI spending and 84% had not fully operationalized AI financial management.
Gartner has separately forecast that an average global Fortune 500 enterprise will have more than 150,000 AI agents in use by 2028, compared with fewer than 15 in 2025. Gartner said only 13% of organizations believe they have the right AI-agent governance in place. Gartner’s research on AI-agent sprawl provides the underlying forecast and governance figures.
Dataiku is entering a market where other major technology companies are building agent inventories and governance systems.
AWS made its Agent Registry generally available on August 31, 2026. The service provides a private catalog and discovery layer for agents, tools, skills, MCP servers and custom resources, with features including approval workflows, audit trails and automatic discovery of agents running on supported AgentCore environments.
Microsoft’s Agent Registry provides tenant-wide visibility across Copilot Studio, pro-code and non-Microsoft agents. Microsoft says the registry can expose agent metadata, usage and performance information and supports administrative actions including blocking, reassigning and deleting agents.
ServiceNow expanded its AI Control Tower in May 2026 with capabilities for discovering, observing, governing, securing and measuring AI systems, agents and workflows across enterprise environments. The company said the expansion added 30 enterprise integrations spanning AWS, Google Cloud, Microsoft Azure, SAP, Oracle and Workday.
Google Cloud also provides an Agent Registry that can automatically register agents hosted on supported Google Cloud runtimes. Agents hosted externally or on unsupported runtimes can be registered manually.
Dataiku’s stated distinction is that Agent Management is designed to sit above those individual platforms rather than serve only one vendor’s agent environment. The company said its inventory is intended to treat agents consistently across connected platforms and provide a portfolio-wide view of performance and risk.
Dataiku is scheduled to demonstrate Agent Management in a live session on October 1, 2026, ahead of its planned general availability later in October.
With the launch, Dataiku is targeting an enterprise problem created by the spread of AI agents across cloud platforms, business applications and custom development environments: maintaining a current record of what agents exist, who owns them, how they perform, what they cost and which ones require additional oversight.
Discover more from Aree Blog
Subscribe now to keep reading and get access to the full archive.
