- Published on
Enterprise AI organizations know how to govern agents but still can't meter what they cost
- Authors
- Name
- DeepHorus
Enterprises are rapidly embracing agentic orchestration, but new research reveals a critical blind spot in cost control. According to the latest VentureBeat Pulse Research survey of 107 large enterprises, the vast majority run multiple orchestration platforms simultaneously, yet one in five still lacks real-time safeguards to halt runaway agents before incurring massive expenses. The findings paint a picture of a maturing yet fragmented market where governance and flexibility take precedence over model loyalty.
The survey, fielded in a single July 2026 wave, targeted organizations with 100 or more employees, with a notably large-enterprise, technology-heavy sample. More than half of respondents come from companies with 10,000 or more employees. This cross-sectional snapshot reveals that orchestration has become plural: 85% of enterprises run two or more orchestration platforms, and 64% run three or more, averaging 3.1 platforms per organization.
Microsoft AI Foundry / Copilot Studio leads in primary usage, present in 70% of stacks, followed closely by OpenAI’s Agents SDK at 68%, and Anthropic’s Claude Platform at 47%. When asked to choose a single primary platform, Microsoft took first place with 41%, while Anthropic secured 28%. No enterprise in the sample relies on just one orchestration layer as a deliberate strategy.
The driving force behind platform selection is flexibility across models and tools, cited by 29% of respondents—nearly three times the 10% who prioritize native alignment with a state-of-the-art base model. Security and permissions (17%), production reliability (15%), and control over agent execution (15%) round out a buying logic focused on governance and optionality rather than developer convenience.
Looking ahead, 53% expect a hybrid control plane by the end of 2026, combining provider-native and external orchestration. The top concern with provider-resident control is security and permissioning limitations (37%), outpacing vendor lock-in at 23%. Consequently, enterprises are prioritizing investment in agent monitoring and debugging (31%) and security enforcement (30%), ahead of workflow tooling at 19%.
Yet the research exposes a significant gap between ambition and reality. A plurality of 47% of respondents admit that only 26–50% of their agents are genuinely orchestrated, while 37% say a quarter or fewer qualify, and just 16% exceed the halfway mark. Many so-called agents remain simple chatbots, despite enterprise aspirations.
Fiscal control emerges as the softest spot in the entire stack. Alarmingly, 21% of enterprises track agent spend only through post-hoc logs, with no real-time ability to stop a runaway execution loop before the invoice arrives. This lack of metering could lead to unexpected cloud bills and governance failures as agent deployments scale.
The findings signal a broader trend toward provider-agnostic architectures in enterprise AI, where the control plane is deliberately hybrid. As organizations invest in visibility and security ahead of convenience, the next challenge will be closing the cost-control gap. Without real-time meters and kill switches, even the most flexible orchestration strategies risk financial and operational derailment.
