Dataiku launches Agent Management to help enterprises track AI agents across platforms

Dataiku Inc. unveiled Agent Management today, a standalone offering that identifies and monitors every AI agent deployed within an organization, even those created by competing vendors. Alongside this release, the company enhanced Cobuild, its natural-language tool for constructing AI initiatives.

Agent Management addresses a critical blind spot for large enterprises. While most organizations maintain detailed records of software ownership and license renewal dates, they struggle to do the same for AI agents. Research from IBM Corp. that Dataiku cited found fewer than one in five organizations maintain a thorough and up-to-date catalog of their AI systems.

"A bank can say how many servers it runs to the decimal," said Florian Douetteau, co-founder and chief executive of Dataiku. "Ask the same bank about its AI agents and the answer is a shrug or a guess."

How Agent Management works

The platform connects to environments where agents operate and consolidates them into a unified registry. Integration points include Salesforce Inc.'s Agentforce, agent services from Amazon Web Services Inc., Microsoft Corp., and Google LLC, as well as systems from Databricks Inc. and Snowflake Inc. Custom environments can integrate through the OpenTelemetry standard, and Dataiku's own agents are supported.

Each agent entry documents the tools and models it relies on, giving supervisors visibility into its operations. High-risk agents receive ongoing documentation of their certification status and identified risks. Scheduled testing creates an audit trail that satisfies regulatory inquiries.

Dataiku positions Agent Management as distinct from monitoring capabilities embedded within individual vendor platforms, which it contends tend to prioritize that vendor's own agents. By sitting above all vendor stacks, Agent Management can answer organization-wide questions such as where risk concentrates or which agents deliver returns below their costs. Users can ask these questions using natural language.

Cobuild expansion

Cobuild, which became generally available in June as an agent that converts business requests into governed Dataiku projects without requiring code, now extends across the broader platform.

Cobuild Insights is a new capability that allows any employee to pose questions about governed company data and receive what Dataiku characterizes as analyst-grade responses. Users can then pursue deeper data exploration within Dataiku using the same line of inquiry.

Dataiku Headless brings Cobuild to developers working in coding agents they already use, including Anthropic PBC's Claude Code, OpenAI Group PBC's Codex, and Space Exploration Technologies Corp.'s Cursor. Developers can construct or modify pipelines, models, and agents directly from these tools.

Migration capabilities allow legacy Microsoft Excel workbooks or Alteryx Inc. workflows to transform into governed projects. The tool can also examine live environments to identify and troubleshoot issues.

Cobuild integrates into Agent Management, enabling teams to request business-value metrics for agents or resolve risk alerts. The expanding AI Catalog now encompasses all approved company assets, including semantic models. Users can state a business requirement and have Cobuild assemble matching approved assets into a solution or visual pipeline with complete lineage.

"Enterprises tend to run technical and business teams on separate platforms without shared data or shared guardrails," said Clément Stenac, Dataiku's co-founder and chief technology officer. "A few pilots could live with that. At scale, it becomes a liability companies have to prepare for."

Availability and pricing

Agent Management will reach general availability in October. The product will charge an annual per-instance fee, with monitoring costs metered by the number of agents. Dataiku has not disclosed timing or pricing for the Cobuild expansions.

CIO survey reveals governance gaps

Dataiku released the "Global AI Confessions Report: CIO Edition, 2026" at its Succeed conference in New York today. The Harris Poll conducted the survey of 685 chief information officers from July 9-29 across eight countries at organizations with more than $500 million in annual revenue.

The findings reveal a significant disconnect between perceived and actual oversight. Nine in 10 respondents reported confidence in their complete agent tracking. However, when asked specifically about agents created outside approved systems or formal channels, 81% acknowledged lacking complete oversight.

Employee-driven agent and application development is outpacing IT governance capabilities, according to 84% of respondents. Most CIOs also believe their career trajectories depend on AI success, and 76% said their positions would face risk if their companies failed to demonstrate measurable AI business returns by the end of 2027. Budget pressure is mounting, with 72% expecting cuts or freezes if this year's performance targets slip.