Dataiku launches agent management amid UK oversight gap

Dataiku has launched Agent Management, a standalone product for tracking and managing AI agents used by enterprises. It is designed to give organisations a single view of agents built on different platforms.

The launch addresses a problem becoming more visible as businesses deploy larger numbers of AI agents across teams and software environments. Many organisations can account in detail for conventional software assets but cannot say with confidence how many AI agents they are running, who owns them or what risks they pose.

Research released alongside the launch suggests the issue is particularly acute in the UK. More than half of British Chief Information Officers, or 53%, said at least one of their AI agents had violated business policy in a way that affected them or their customers, compared with a global average of 31%.

That finding points to a widening gap between adoption and oversight. As companies roll out agents for customer service, internal workflows and operational tasks, they are also creating a more complex estate of models, tools and software dependencies that may span several vendors.

Agent Management is intended to address that by discovering agents across an organisation regardless of where they were built. It connects with platforms including AWS Bedrock, Databricks Agents, Google Vertex, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, Snowflake Cortex and Dataiku's own software, while also supporting custom environments through OpenTelemetry.

Once connected, the product builds a central inventory of agents and identifies the structure behind them, including the models and tools they use. That is meant to provide visibility not only into an agent's existence but also into how it operates.

Oversight gap

The broader backdrop is a market moving quickly towards agent-based AI tools, with governance still catching up. Dataiku cited IBM research showing that fewer than one in five organisations keep a complete and current inventory of their AI systems.

Most existing monitoring tools are tied to a single provider's platform, which means businesses using several vendors can end up with fragmented oversight. Dataiku argues that an independent layer sitting above those platforms can help companies identify which agents are not being monitored, where risk is concentrated and which systems are delivering value relative to their cost.

For higher-risk agents, such as those dealing with customers, sensitive data or live transactions, the product keeps records covering certification status, named risks and scheduled tests. This creates an evidence trail that can be reviewed by managers, auditors or regulators.

The commercial model reflects how companies are likely to use the software as a control tool rather than just a development product. Access is priced annually per instance, while monitoring is metered per agent.

The launch also adds to a growing contest among software providers seeking to become the management layer for enterprise AI rather than only the environment in which models or agents are built. By focusing on multi-vendor estates, Dataiku is trying to position itself where enterprises need visibility across competing tools already in use.

Executive view

Florian Douetteau, Co-Founder and Chief Executive Officer of Dataiku, framed the issue as one of basic operational control.

"Ask a bank how many servers it runs, and you get an answer to the decimal. Ask how many AI agents it's running, and you get a shrug or a guess," said Florian Douetteau, Co-Founder and Chief Executive Officer of Dataiku.

He said the lack of oversight was not a deliberate design choice but the result of deployment moving faster than inventory practices and controls.

"Nobody set out to build it this way. Teams built agents faster than anyone could count them. Agent Management tells you what's actually out there, and what it's actually worth," Douetteau said.

The product enters the market as businesses face increasing scrutiny over the use of AI in customer interactions and business operations. That scrutiny is likely to focus not only on whether AI systems work as intended, but also on whether companies can demonstrate who approved them, what data they use and how risks are being monitored.

For UK businesses, the figures cited by Dataiku suggest the governance challenge is no longer theoretical. If more than half of technology leaders have already seen an agent breach policy with customer or business impact, the task may now shift from experimentation to control.

Dataiku said Agent Management provides a whole-portfolio view so teams can ask questions in plain language and receive answers covering the full estate of AI agents.