VentureBeat Pulse’s June 2026 survey of 101 enterprises (100+ employees) shows rapid consolidation of agent orchestration onto major model-provider platforms — with Anthropic’s Claude far out front — while revealing a large gap between orchestration plans and what deployed agents actually do. Companies are buying platforms and tooling for reliable multi-step execution and prefer hybrid control planes to avoid vendor lock-in, yet the majority of deployed agents remain single-prompt chatbot wrappers and many organizations lack real-time cost controls.
Key figures and context
- Among 101 respondents, 40% named Anthropic’s Claude as their primary orchestration platform, compared with Microsoft (18%) and OpenAI (13%). The five major providers (Anthropic, Microsoft, OpenAI, Google, Amazon) together account for roughly 80% of primary deployments. Open frameworks (e.g., LangChain/LangGraph) and in-house builds are single-digit shares.
- Platform satisfaction averages 3.94/5 (109 responses); ease of implementation is weakest at 3.85. Ninety-six percent of users expect to change their orchestration approach within a year.
- The single largest factor driving platform choice is the pull of the underlying model — “model gravity” (21%). Flexibility across models/tools and ease of development each account for 17%.
- Success metrics are reliability-focused: task completion reliability (32%) and multi-step workflow management (28%) together represent 59% of responses.
The chatbot trap: ambitions outrun deployed reality
When asked to assess their portfolios, 71% of enterprises said a quarter or fewer of their deployed “agents” are genuine multi-step orchestrations; only 10% reported that over half their agents are multi-step. In short, the orchestration layer (platforms, control plane designs, budgets) is being built ahead of the orchestrated portfolio it’s intended to run.
Smaller organizations are more likely to be in this trap: 77% of firms under 2,500 employees report that a quarter or fewer of their agents perform true multi-step work, versus 62% for larger firms.
Near-term strategic moves
Over the next 12 months the three most commonly anticipated changes are tightly grouped: building an in-house control plane (25%), standardizing on a single framework (24%), and moving agents from sandbox to production (23%). These choices indicate a shift from experimentation toward consolidation and operationalization.
Sixty-eight percent plan to adopt a new, additional, or replacement orchestration platform within a year; the largest group among those actively shopping (29% of all respondents) has no shortlist yet. Among named contenders for those considering change, OpenAI leads at 16%, LangChain/LangGraph at 12%, and Anthropic at 7%.
Where investment is going
Budget growth is concentrated on agent workflow tooling (34%), then security and permissions enforcement (25%), and scaling infrastructure (20%). Monitoring and debugging drew 11%.
This spending mix aligns with the priority on reliable multi-step execution: companies are buying the machinery to string steps together and to secure production deployments rather than merely to observe them.
Control plane preference and lock-in fears
A majority (51%) expect the primary control plane to be hybrid by the end of 2026 (provider-native plus external orchestration). Only 6% expect to hand full control to a provider-managed service. Overall, 88% prefer architectures that keep at least part of control outside the model provider.
Vendor lock-in is the top worry (35%), followed by security/permissioning limitations (28%) and inflexibility across models/tools (21%). Compared with an April–May survey wave, concern about lock-in has risen relative to security, indicating a shift from worrying whether platforms can be secured to worrying whether they can be replaced.
Fiscal control remains reactive for many
Regarding control over token consumption and runaway agent cost, 27% have no real-time programmatic way to stop an agent before a bill arrives and learn only from logs afterward. Thirty-two percent rely solely on native platform caps and throttles. Custom gateways are used by 23% and cross-model routing for cost arbitrage by 19%.
Smaller enterprises are more likely to have only reactive spend control (about 34% for firms under 2,500 employees versus 20% for larger firms).
Bottom line
Enterprises with 100+ employees are standardizing quickly on major model-provider platforms — Anthropic’s Claude leads the cohort — and define orchestration success by reliable multi-step completion. They are investing in workflow tooling and security, planning hybrid control planes to avoid vendor lock-in, and preparing to push agents into production. But measured honestly, most deployed agents are still simple chatbot wrappers, and meaningful real-time fiscal controls are missing in many organizations. The orchestration layer is being shaped well before most agents perform the workflows that layer is built to manage. The coming months and waves of research will show how quickly deployed reality closes the gap on these ambitions.
Methodology
VentureBeat fielded this Pulse Research survey in June 2026, sampling 101 organizations with 100 or more employees. The cross-sectional, self-selected sample includes product and program managers, CIO/CTO/CISO roles, consultants, and directors/VPs of data, AI, and engineering across Technology/Software (44%), Financial Services (17%), Healthcare/Life Sciences (8%), and other sectors. Results should be read directionally rather than as precise market-share measures.
