Dataiku launched Agent Management in September. Oracle devoted a chunk of its AI World conference to the same topic. Companies deploying AI agents at scale are discovering that visibility into what those agents actually do is harder than building the agents in the first place.
The technical challenge is real. An AI agent that queries customer data, writes to a CRM and triggers downstream workflows creates an audit trail that sprawls across systems. Logging each step is one thing. Understanding whether the agent made the right call, with the right data, under the right policy constraints, requires a different kind of oversight. The vendors now racing into this space are betting that centralised agent governance becomes as essential as data lineage tools became five years ago.
For our Data & Digital clients in the DACH region, this is already shaping hiring. A manufacturing company in Stuttgart recently asked us to find someone who could own agent observability. The spec called for a professional who understood model behaviour, data access policies and production monitoring well enough to work across all three domains. Similar requests have come through from financial services and logistics companies over the past quarter.
The candidates best positioned for these roles have a data platform background, often Databricks or Snowflake, combined with hands-on experience of MLOps pipelines. They can read a model’s decision log and translate what it means for data governance. They present findings to senior stakeholders in plain terms.
Our advice to data leaders: treat agent governance as a hiring priority now. Regulators will eventually ask harder questions about automated decision-making, and internal capability takes time to build.
Prompted by reporting from Datanami.