Agentic orchestration: Enterprise AI organizations know how to govern agents but still can't meter what they cost

Across 107 enterprises, it is not a choice of a single platform that enterprises opt for agentic orchestration. The typical enterprise runs three orchestration platforms simultaneously, selecting them for flexibility across models rather than preference for any one in particular. Microsoft emerges as the leading choice with 41% of respondents, followed by Anthropic at 28%.

No respondent in the sample opted for a singular orchestration layer and declared it their strategy. The selection process is driven by flexibility across models and tools, which is the primary purchase driver at 29%, significantly higher than any other factor. Security and permissions (17%), production reliability (15%), and control over agent execution (15%) are the key considerations, reflecting a focus on governance and optionality.

A majority (53%) anticipate a hybrid control plane by the end of 2026, combining provider-native and external orchestration. Security and permissioning limitations (37%) are the primary concern regarding provider-resident control, surpassing vendor lock-in (23%) and limited visibility (22%). Investment has shifted towards agent monitoring and debugging, with 31% of enterprises allocating resources to it, followed by security and permissions enforcement at 30%.

Most companies admit that their "agents" are primarily chatbots, with 47% saying between 26 and 50% of their agents are genuinely orchestrated. However, fiscal control remains an issue, as 21% of enterprises only track agent spend through post-hoc logs, lacking real-time control mechanisms. The survey, part of VentureBeat's Pulse Research series, was conducted in July 2026, focusing on enterprises with 100 or more employees.

The sample is large, technology-focused, and predominantly composed of software and ML engineers, product and program managers, directors of data/AI/analytics, and VPs of data/AI/analytics.