Collibra’s new AI Command Center promises to combat agentic hallucinations

Big data management startup Collibra BV is trying to position itself as the nerve center for artificial intelligence agents with the launch of its new AI Command Center offering.

It’s designed to give enterprises a simple way to orchestrate fleets of governed AI agents in real time with minimal hassle. With Collibra’s AI Co/mmand Center, companies will be able to access a proactive, continuous lifecycle management plane that treats each AI agent as a dynamic entity that requires constant supervision, the company explained.

It’s launching at a critical juncture in the emerging “agentic AI” industry. Whereas the first wave of the AI boom was all about generative AI chatbots, most companies are now pushing more advanced systems that can work autonomously to perform tasks on behalf of human users with minimal supervision. Updating customer records, processing transactions, managing call centers and so on are increasingly being automated in this way, but they don’t always go as intended.

As Collibra Chief Executive Felix van Van de Maele explains, almost every enterprise that’s trying to use AI agents has to pay a kind of “hallucination tax,” which refers to the hidden costs of manual oversight, correcting agents’ mistakes and the general risk that things might go wrong. “The AI Command Center eliminates that tax,” he promised. “It gives organizations real-time visibility, continuous control and the confidence to run AI at the speed it actually moves.”

Collibra highlighted the scale of this “hallucination tax” in a recent survey. It found that while 91% of enterprise technology decision-makers are experimenting with AI agents, less than half of them have successfully established the governance policies needed to oversee them.

That means there’s a huge number of AI agents up and running in production that lack clear ownership and traceability, and so if things go wrong, companies will very often not be able to understand which agent was responsible, or how it messed up. That leaves them exposed to massive risks, both in terms of reputation and regulation.

Collibra’s AI Command Center is meant to fix this. It’s being launched through a strategic partnership with the large language model testing startup Giskard AI SAS, and is meant to function as a centralized dashboard through which teams can see, monitor and control every AI agent that has been deployed.

The company is promising real-time visibility with insights into the ownership, behavior and reasoning behind the decisions of each AI agent. It also detects agent drift, which is the term used to describe when agentic systems begin to “drift” from their intended parameters. By catching them early, humans can intervene before disasters strike. In addition, the AI Command Center also provides companies with access to out-of-the-box templates for evaluating AI agent’s readiness and risk in-line with AI UC-1 compliance standards.

As for Giskard, it enables execution-level directly at the governance layer, meaning that AI engineers can perform extensive testing while their agents are still in the development pipeline and obtain feedback. “Enterprise AI leaders need both top-down governance on how agents are built, and bottom-up tests that operationalize this governance directly inside the CI/CD pipelines of AI engineers,” said Giskard co-founder and co-CEO Alex Combessie. “That’s exactly what Giskard and Collibra now deliver together: governance that actually reaches the execution layer, where AI risks really live.

Collibra’s platform is fully integrated with the Model Context Protocol. It offers an MCP Server that enables it to deliver governed metadata and business context directly to agents in real time. This means that whenever agents take actions, they’re doing so based on trusted and structured corporate data, rather than outdated information or things they’ve dreamed up by themselves.

The biggest challenge will be integration. The enterprises it’s targeting rely on a sprawling mess of third-party LLMs, vector databases and AI orchestration frameworks, and it will need to find a way to unify all of these systems if it’s to avoid becoming just another dashboard that developers ignore. Recognizing that, the company has also joined Databricks Inc.’s new “Agent Bricks” ecosystem as a founding member to ensure its platform plays nicely with key components of the agentic AI stack.

Image: SiliconANGLE/Microsoft Designer