“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.” As Florian Douetteau, CEO of Dataiku, recently put it, this sentiment captures the core tension in modern IT: we are rapidly deploying autonomous systems without the foundational inventory or governance frameworks required to manage them.
The data confirms this disconnect. According to a March 2026 Cisco survey, while 85% of organizations are piloting or deploying agentic AI, only 5% have reached broad production. Security remains the primary barrier, cited by 60% of respondents. It is worth noting that this is a vendor-commissioned study, yet its findings align with broader industry trends regarding the “agent trust gap.”
This gap is not merely a technical hurdle; it is a structural one. Gartner forecasts that over 40% of agentic AI projects will be canceled by the end of 2027. When we layer in the June 2026 VentureBeat Pulse Research, which found that 50% of enterprises shipped agents that passed internal evaluations but still caused customer-facing failures, the picture becomes clear: we have a reality-alignment problem, not a coverage problem.
The industry is responding with a flurry of governance products. In just the last few weeks, we have seen a surge in tooling designed to bring order to this chaos. SAP launched its AI Agent Hub in August, focusing on vendor-agnostic inventory. Collibra introduced Guardian Agents in September to provide runtime supervision. Dataiku announced the general availability of its Agent Management platform for October. Island pivoted its browser-based security model to an agentic control plane, and Microsoft unveiled its Copilot Autopilot with integrated Entra identity governance.
These tools are necessary because the measurement gap is widening. A Collibra-commissioned Harris Poll from September 2026 found that 76% of decision-makers face critical roadblocks moving agents from pilot to production. While we must treat vendor-commissioned data with a degree of skepticism, the convergence of these findings-from Cisco, Gartner, and VentureBeat-suggests a systemic issue. The industry is struggling to move from the “shrug or guess” phase to a state of operational maturity.
Ownership fragmentation further complicates the issue. The Cisco research highlights that governance responsibility is split between CISOs (29%), CIOs (27%), and AI committees (24%), with 11% of organizations reporting no clear ownership at all. As SAP CTO Philipp Herzig noted, “Agents are everywhere. Some are great, some are not, and almost no one has a consistent picture-no central governance, no clear view of what each agent does.”
This lack of a central view is exactly what the new wave of governance tools aims to solve. Whether it is through SAP’s inventory approach, Collibra’s runtime contracts, or Island’s control plane, the goal is to move AI from a “black box” experiment to a managed enterprise asset. As Island CTO Dan Amiga observed, “Agents do not operate in a single layer of the technology stack, so they cannot be governed from one.”
For those tracking these developments, our previous coverage provides essential context on SAP’s inventory approach, Dataiku’s control layer, and NiCE’s routing layer.
The next 18 months will determine which AI projects actually deliver value. Success will depend on moving past the hype to implement real, runtime-level governance. Organizations that want to scale AI must prioritize control now, before the costs of unmanaged agents become unsustainable.