Dataiku Japan announced Agent Management on October 14, 2026, a cloud service that automatically discovers and centrally manages AI agents running across multiple services. It is scheduled to launch on October 15, 2026. The service lists agents, monitors their operations and costs, and assesses risks; based on those results, it can send alerts or connect with IT service management products. It can be used on its own, without other Dataiku products.

The service connects to each AI agent platform’s API to collect logs and telemetry (data showing system activity). Agents developed in-house or run on premises can connect through OpenTelemetry, a standard for emitting measurement data in a shared format. Supported platforms are Google Vertex, AWS Bedrock, Microsoft Copilot Studio, Azure Foundry, Dataiku, Snowflake Cortex, Databricks Agents and Salesforce Agentforce.

Monitoring covers operations, usage, costs and business value. Costs are calculated from token use and models, with the option to apply company-specific formulas. Users can aggregate costs by service, model or agent to identify where spending is accumulating. Business value is measured using KPIs set for each agent. In a demo, a factory agent showed $20,000 in savings from preventing downtime over the past 30 days. The dashboard also showed that nearly 19% of its work instructions were unsafe, above the 10% threshold for a critical rating.

For risk management, the service scores each agent by the potential impact and likelihood of risks, then uses a risk matrix to prioritize responses. It can also group risks by type, alert users about high-risk agents and connect with IT service management products. Dataiku Japan cited “rogue AI”—agents created by frontline teams without IT’s awareness—as a reason for the launch, calling it “a repeat of problems seen with Excel macros, shadow IT and RPA.”

Dataiku Japan announced Agent Management on October 14, 2026, a cloud service that automatically discovers and centrally manages AI agents running across multiple services. It is scheduled to launch on October 15, 2026. The service lists agents, monitors their operations and costs, and assesses risks; based on those results, it can send alerts or connect with IT service management products. It can be used on its own, without other Dataiku products.

The service connects to each AI agent platform’s API to collect logs and telemetry (data showing system activity). Agents developed in-house or run on premises can connect through OpenTelemetry, a standard for emitting measurement data in a shared format. Supported platforms are Google Vertex, AWS Bedrock, Microsoft Copilot Studio, Azure Foundry, Dataiku, Snowflake Cortex, Databricks Agents and Salesforce Agentforce.

Monitoring covers operations, usage, costs and business value. Costs are calculated from token use and models, with the option to apply company-specific formulas. Users can aggregate costs by service, model or agent to identify where spending is accumulating. Business value is measured using KPIs set for each agent. In a demo, a factory agent showed $20,000 in savings from preventing downtime over the past 30 days. The dashboard also showed that nearly 19% of its work instructions were unsafe, above the 10% threshold for a critical rating.

For risk management, the service scores each agent by the potential impact and likelihood of risks, then uses a risk matrix to prioritize responses. It can also group risks by type, alert users about high-risk agents and connect with IT service management products. Dataiku Japan cited “rogue AI”—agents created by frontline teams without IT’s awareness—as a reason for the launch, calling it “a repeat of problems seen with Excel macros, shadow IT and RPA.”