Enterprise AI agents: Ambition outpaces reality, study finds
A new VentureBeat Pulse survey of 101 enterprises reveals that most deployed 'agents' are still chatbot wrappers, while only a few use true multi-step orchestration, despite rapid platform consolidation. Anthropic's Claude leads with 40% primary platform share.
Agentic orchestration ambitions vs. reality
The VentureBeat Pulse Research, surveying 101 enterprises, reveals a significant gap between enterprise ambitions and actual agent deployment. While organizations are quickly consolidating onto major model platforms, the majority of deployed 'agents' remain simple chatbot wrappers rather than true multi-step orchestrated workflows.
Platform choice and leaders
Anthropic's Claude is the primary platform for 40% of respondents, more than double its nearest competitor Microsoft (18%) and OpenAI (13%). The choice is driven primarily by 'model gravity'—21% cite native alignment with a state-of-the-art base model. Success is measured by reliable multi-step execution: task completion reliability (32%) and multi-step workflow management (28%) are the top metrics.
Reality: most 'agents' are not real agents
Despite high ambitions, the reality is different. 71% of respondents admit that a quarter or fewer of their deployed 'agents' are true multi-step orchestrated workflows rather than single-prompt chatbot wrappers. Only 10% of enterprises have crossed the halfway mark. The orchestration layer is being built well ahead of the orchestrated portfolio it is meant to run.
Future architecture: hybrid control plane
By the end of 2026, 51% of enterprises expect a hybrid control plane—combining provider-native and external orchestration. Only 6% plan to hand full control to a provider-managed service, because vendor lock-in (35%) is the most feared risk if control resides inside a single provider.
'The core finding is a gap between orchestration ambition and orchestration reality,' the report states. Enterprises have a deployment problem, not a platform problem—the technology is available, but effective use still lags.
