Enterprise agent orchestration: Ambition outpaces reality as most 'agents' remain chatbot wrappers

Enterprise agent orchestration platform choices are highly concentrated among a few model providers. The survey shows Anthropic Claude leads with 40% share, more than double that of second-place Microsoft (18%), with OpenAI at 13% in third. Google, Amazon, and others together account for about 80% of deployments, while open-source frameworks like LangChain and in-house systems hold only single-digit shares.

This concentration trend aligns with the logic of 'model gravity': enterprises tend to choose orchestration environments closest to their standardized frontier models. However, it's worth noting that 3% of enterprises have not yet implemented any orchestration, indicating the market is still in early adoption.

The top factor in platform selection is model gravity (21%), i.e., native alignment with advanced foundation models. However, flexibility and ease of development (each 17%) closely follow, indicating that enterprises also want to avoid being locked into a single choice, setting the stage for future lock-in concerns.

Security and permissions (14%) and total cost of ownership (11%) are also important considerations, while performance (latency/memory) accounts for only 4%, suggesting that current constraints lie in model fit and availability rather than raw speed.

The primary metrics enterprises use to evaluate orchestration success are task completion reliability (32%) and multi-step workflow management (28%), together accounting for 59%. Developer productivity (17%) and end-user experience (9%) are relatively less important, indicating that orchestration is viewed as an internal execution issue rather than a user experience problem.

This reliability-first criterion contrasts sharply with the 'chatbot trap': enterprises define success as reliable multi-step execution, but most deployed agents have not yet achieved multi-step work.

The survey reveals a huge gap between orchestration ambitions and reality: 71% of enterprises say that only a quarter or less of their deployed agents are true multi-step workflows, rather than single-prompt chatbot wrappers. Only 10% of enterprises cross the halfway mark.

This gap varies by enterprise size: 77% of small enterprises admit that no more than a quarter of their agents are true multi-step workflows, indicating that smaller organizations are more prone to falling into the 'chatbot trap.'

By the end of 2026, 51% of enterprises expect to adopt hybrid control planes (provider-native plus external orchestration), with only 6% willing to hand over full control to providers. Concerns about vendor lock-in (35%) are the primary driver behind this choice.

Investment directions align with this: agent workflow tools account for the largest investment share (34%), followed by security and permission enforcement (25%). This indicates that enterprises are investing in orchestration layer construction, even though their orchestration portfolios are not yet mature.

Financial control is a clear weakness in enterprise orchestration: 27% of enterprises have no real-time means to stop runaway agents from generating unexpected bills. This gap contradicts the reliability-first standard of orchestration, as the inability to control costs may undermine trust in agents.

The survey also shows that users rate current platforms at 3.94/5 (out of 5) satisfaction, with 'ease of implementation' scoring the lowest (3.85). 96% of users plan to change their orchestration approach within a year, indicating that current platforms are only 'temporarily accepted,' not a long-term satisfaction.