Three questions, three different races
VentureBeat's headline is accurate on the question it chose. Among companies running an agent orchestration platform, OpenAI's is used by roughly twice as many as Anthropic's: 43% to 22%, with Google close behind Anthropic at 21%. Orchestration is the layer that hands tasks to agents, calls tools and holds their permissions, so this is the closest thing the enterprise agent market has to a share figure.
Ask the same respondents which platform is their primary one, and the race changes shape. Primary platform, among the same 107 respondents: OpenAI 25% (27), Google 18% (19), Anthropic 16% (17). OpenAI leads Anthropic by nine percentage points, compared with a 21-point gap in the multiple-selection use question. The gap between the two answers is the distance between trying a platform and building on it, and a use figure counts both identically.
The survey also found that 73% of platform users reported using just one orchestration platform, while 27% used two or more. The primary-platform question therefore measures preference across a wider field, not merely a choice among OpenAI, Anthropic and Google.
Ask a third question, which platforms companies are considering for the next year, and the order flips. Among 61 respondents who named at least one candidate, 33% (20) included Anthropic, 30% (18) OpenAI and 25% (15) Google. Anthropic is numerically first, but VentureBeat says the differences are not statistically significant. Of the 20 considering Anthropic, 12 do not currently use it. That group is small, and VentureBeat flags the size itself. Small or not, it points the opposite way from the headline.
The payment data measures something else again
None of this contradicts Ramp's widely shared figures, which show more businesses paying Anthropic than OpenAI. Ramp's September 9 report covers August payment activity among businesses using its platform. It records whether a business paid each provider, not which orchestration platform it uses or how much of its AI workload runs there. Ramp counts paying businesses on its own platform, each counted once regardless of size, in a customer base VentureBeat itself describes as more tech-forward than average. Its survey counts people at organisations of 100 or more employees, plus consultants answering for a main client. A company can pay for Claude subscriptions and run its agents on OpenAI's platform in the same week; large retailers in the survey describe exactly that kind of split, with one vendor's models writing code and another's reviewing it. Use, primary use and payment are three measurements. Treating any one of them as "the market" is where the error enters.
The number with a budget attached
The survey's most practical finding is not about share at all. Nearly two-thirds of companies had replaced the model inside a production app in the past year. Of those, 59% saw performance drop and had to rework the app around the new model or revert, and only 19% described the swap as taking almost no engineering work. Consultants reported swapping at far higher rates than in-house respondents (81% against 51%), which pushes the overall figure up, but even the lower in-house rate means swapping is a normal, repeated cost rather than an edge case.
That reframes the platform question. If a model swap usually costs days of retesting prompts, tools and evaluations, then the orchestration layer is less a crown to win than a commitment to price in. A lead measured in trial use can move quickly. A lead measured in primary deployments, where the rework bill lands, moves slowly, and on that measure nobody in this survey is running away with anything.
What would settle it
VentureBeat plans to repeat the questions in its October survey, with a fresh group of respondents. That repetition matters more than any single wave: each month's sample is different, the publisher says so plainly, and September's screening differed from August's, which limits comparisons across waves. A similar result would strengthen confidence that the pattern is repeatable within its survey population, but it would not by itself establish representative enterprise market share or a structural advantage. Until then, the honest summary is the duller one: OpenAI is ahead on every measure in this survey, by margins that range from commanding to narrow depending on which question you pick.
Sources
- [1] VentureBeat Intelligence — September 2026 VB Pulse Agentic Orchestration Report
- [2] VentureBeat, "OpenAI's AI agent platform is used by twice as many companies as Anthropic's, new VentureBeat survey finds," 13 October 2026Read source
- [3] Ramp — September 9 AI Index (August payment activity), figures as reported within the VentureBeat article