.png?w=1024)
A company may know where its AI agents were built and still struggle to answer a simpler question: Which ones are running now?
Dataiku announced Agent Management on September 24, saying the standalone product will bring agents built on different platforms into one inventory. For enterprise teams deploying agents across cloud services and business applications, the appeal is a shared view of ownership, performance, cost, and risk. Dataiku says the product will be generally available in October 2026.
That timing matters. Buyers can review Dataiku’s stated integrations and request a demo now. Whether it finds and monitors agents consistently across their own environments will need to be tested when the product becomes available.
What Agent Management is designed to show
According to Dataiku’s announcement, Agent Management scans connected platforms to create an agent inventory. Dataiku says it can identify the tools and models an agent uses, while its product page describes records of each agent’s owner, purpose, usage, cost, and quality. For higher-risk agents, the company also promises documented risk assessments, certification status, and tests that run on a schedule.
Dataiku names AWS Bedrock, Databricks, Google Vertex, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, Snowflake Cortex, and its own platform among the environments it plans to connect. Its product page also lists n8n, while the announcement describes OpenTelemetry support for custom environments.
The proposed inventory addresses a governance problem: records are difficult to keep current when separate teams deploy agents on separate platforms. As eWeek reported in its coverage of the enterprise AI governance gap, visibility and traceability become more consequential as AI systems move into production and gain access to business workflows.
What eWeek found: Cross-platform visibility depends on each integration
Dataiku’s announcement promises a single view of agents across platforms. Its product documentation, however, says Agent Management connects through APIs and log streams, and that the depth of each integration depends on what the connected platform exposes. Dataiku says deeper integrations are still being developed.
That qualification has a practical consequence: an agent appearing in the inventory does not, by itself, establish that a team can see the same activity details and risk evidence for that agent as it can for agents on another platform. This is a finding from comparing Dataiku’s published claims and documentation; eWeek has not tested the product.
The company also says Agent Management is not an AI gateway. It sits above agents to track them; it does not sit between an agent and its tools to control traffic. Teams evaluating it should distinguish visibility into an agent’s behavior from the ability to stop an action as it happens. That distinction matters when agents can make changes in enterprise systems, a risk explored in eWeek’s report on autonomous AI workflows.
Dataiku describes annual pricing per instance, with monitoring metered per agent, but gives no dollar amount in its announcement. For buyers, the practical questions will be how many agents each connector discovers, what evidence it captures, and how monitoring costs change as the inventory grows. Those checks would also help buyers place it alongside the broader AI governance tools eWeek has compared.
When the product becomes available in October, buyers should test each connector against their own agents: which agents appear, what activity and risk evidence is visible, and what monitoring costs as coverage grows. A shared inventory is useful only if teams understand where its visibility ends
Read more: OpenAI’s six disclosed AI misalignment cases show why teams also need controls over agent credentials, network access, and actions beyond a monitoring dashboard.