Dataiku to Launch 'Agent Management' for Unified AI Agent Oversight

Dataiku said on the 28th that it plans to officially launch 'Agent Management' in October, a feature that lets companies manage AI agents deployed across multiple internal platforms in one place. The feature is designed to track and manage AI agents operating across multiple platforms within an enterprise.

Dataiku said it focused on enabling unified management of corporate AI agents regardless of the deployment platform. The management items include each agent's operating status, performance, cost, and risk level. It also supports applying a consistent management framework across multi-platform environments.

As corporate adoption of AI agents increases, so does the challenge of understanding overall operations. According to IBM's 'AI in Motion' study, fewer than 20% of organizations keep track of all operating AI systems and manage them in a current state.

For companies using multiple AI platforms, management complexity increases. Because the information provided by individual platforms centers on agents deployed within their own environments, it is difficult to consistently identify the people in charge, intended use, costs, and risk factors across the entire company.

To address this, Agent Management provides a function that consolidates AI agents distributed across multiple platforms into a single list. The feature is built on an architecture that connects multiple platforms at a higher management layer, and it is not tied to any specific vendor.

Integration targets include Dataiku's own platform, Amazon Web Services (AWS) Bedrock, Databricks Agents, Google Vertex, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, and Snowflake Cortex. Custom environments are supported through OpenTelemetry.

After integration, users can check the models and tools used by each agent. The scope of visibility goes beyond simply confirming whether an agent exists, allowing users to trace its actual configuration and how it operates.

High-risk agents are managed separately. The conditions for management include customer-facing service delivery, processing sensitive information, and executing real-time transactions. For these agents, authentication status, identified risk factors, and periodic test results are continuously recorded.

The accumulated information is used by internal administrators and can also respond to verification requests from auditors or regulators. This supports a unified management framework and enables operational visibility, risk management, and audit response.

Within the same framework, users can also review each agent's cost and performance together. Costs and performance can be compared across agents, and areas of concentrated risk can also be identified. Based on this, the system supports adjustments to operational priorities.

Managers can perform natural-language queries based on the overall operating status. This makes it possible to identify unmonitored agents and check agents with high risk levels. It also allows users to identify agents with low performance relative to cost.

Dataiku said that, unlike existing multi-agent monitoring tools that operate within a specific vendor's technology stack, its Agent Management was designed for a multivendor environment. As a result, it said a single management framework can be applied even to companies using multiple platforms.

Florian Douetteau, cofounder and CEO of Dataiku, said that as rapid AI agent deployment by individual teams spreads, it is becoming harder to understand the company's overall status. He added that Agent Management helps provide visibility into the status of agents in operation and the business value of each agent.

Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407440

References

• https://www.epnc.co.kr/news/articleView.html?idxno=407440