VentureBeat Pulse research conducted in July 2026 surveyed 107 enterprise respondents (organisations with 100+ employees) about their use of agent orchestration platforms. The central finding is that enterprise orchestration is plural: companies typically run several orchestration platforms in parallel, choosing them for cross‑model flexibility and governance rather than allegiance to any single provider. The average number of platforms per organisation is 3.1.
Which platforms appear and which are treated as "primary"?
- 85% of respondents run two or more orchestration platforms; 64% run three or more. The median organisation runs three platforms.
- Platform presence across the sample: Microsoft AI Foundry / Copilot Studio appears in 70% of stacks, OpenAI Agents SDK in 68%, and Anthropic Claude Platform in 47%.
- When respondents were asked to name a single primary platform (a question that produced one unambiguous choice from 61 respondents), Microsoft led with 41%, Anthropic 28%, LangChain / LangGraph 10%, and OpenAI Agents SDK 7%.
- Average ratings: overall satisfaction 4.17/5, ease of implementation 3.91/5, and value for money 3.63/5 — indicating notable dissatisfaction around cost.
Why organisations choose the platforms they do
- The leading purchase driver is flexibility across models and tools (29%). Organisations prefer platforms that do not lock them into a single base model.
- Model gravity — picking the orchestration layer tied to a preferred frontier model — accounts for only 10%.
- Other important factors include security and permissions (17%), production reliability (15%), and control over agent execution (15%). Ease of development (8%) and total cost of ownership (4%) were lower on the list.
What enterprises optimise for in orchestration
- The primary success metrics are task completion reliability (30%) and multi-step workflow management (27%), which together make up 57% of responses.
- Developer productivity accounts for 23%, while end-user experience is only 7%, underlining that orchestration is judged mainly as an internal execution capability.
Plans to change platforms and vendor consideration
- 67% of enterprises plan to adopt a new, additional, or replacement orchestration platform within 12 months; the largest timing cohort expects to act in 6–12 months (28%), and only 15% plan to move within a quarter.
- Among organisations actively evaluating platforms (72 respondents), Anthropic tops forward consideration at 43%, followed by Google and custom in‑house solutions (each 31%), OpenAI (25%), LangChain / LangGraph (17%), and Microsoft (17%). This contrasts with Microsoft’s strong installed presence, indicating the installed base and the forward pipeline point to different vendors.
- A notable share of enterprises are considering building in‑house control planes rather than buying.
Investment priorities: observability and governance lead
- Planned investment growth is focused on monitoring and debugging (31%) and security and permissions enforcement (30%), which together represent 61% of expected increases.
- Workflow tooling (19%) and scaling infrastructure (18%) trail behind. Only 3% reported a flat budget.
- In short, spending is aimed at seeing and governing agents in production rather than merely building them.
Control-plane expectations and provider risks
- 53% of respondents expect a hybrid control plane by the end of 2026 (a combination of provider-native and external orchestration).
- Overall, 78% favour architectures that keep at least part of control outside of the model provider (hybrid, in‑house, or externally-abstracted solutions) versus 14% willing to hand full control to a provider-managed service.
- The leading concern about provider‑resident control is security and permissioning limits (37%), followed by vendor lock‑in (23%) and limited visibility/observability (22%). Combined, security and visibility-related concerns are more common than purely commercial worries.
Portfolio maturity: chatbot prevalence is decreasing
- 47% of respondents estimate that between 26% and 50% of their deployed agents are genuinely orchestrated multi-step workflows; 16% say more than half are genuinely orchestrated.
- 37% report a quarter or less are truly orchestrated, and pure-chatbot portfolios are now rare (3%).
- Organisations with more orchestrated agents tend to run more platforms: those reporting 26–50% orchestrated agents run an average of 3.5 platforms, while the least orchestrated groups run 2.8 platforms on average.
- Organisation size does not appear to explain portfolio maturity in this wave.
Fiscal control remains a weakness for many
- 21% of enterprises only discover a runaway agent’s cost after the fact through logs — they have no real‑time programmatic way to stop execution before a budget is breached.
- 30% rely solely on provider‑native caps and throttles.
- The remainder either built custom gateways (25%) or use cross‑model routing to control cost deterministically (24%). These deterministic approaches work in part because organisations run multiple platforms simultaneously.
- The lack of robust, real‑time fiscal controls aligns with the weaker value‑for‑money ratings the platforms receive.
Methodology and caveats
The results come from a single July 2026 wave of VentureBeat Pulse Research limited to organisations with 100+ employees (n=107). The sample skews toward large enterprises and technology firms; respondents were largely technical roles (software/ML engineers, product/program managers, directors/VPs of data/AI/analytics). Because the sample is self‑selected and cross‑sectional, figures should be read as directional rather than exact market measures. Where respondents could select multiple answers, percentages may sum to more than 100%.
Conclusion
Enterprises are treating orchestration as a portfolio problem: they run multiple platforms to preserve model choice and governance options, judge success by reliable multi‑step execution, and plan to keep control at least partly outside model providers. Investment is prioritised toward observability and permissions enforcement, but fiscal controls lag — one in five organisations cannot stop a runaway agent in real time. The current footprint reflects where enterprises are today, not where they intend to remain: a mix of vendor presence, in‑house solutions and growing interest in Anthropic and custom control planes may reshape deployments over the next year.
