Claude wins 40% of enterprise agent orchestration bets, but 71% call most “agents” chatbot wrappers

If you’re an enterprise betting on “agentic orchestration,” the uncomfortable truth is this: the orchestration layer is ahead of the orchestrated work. In VentureBeat Pulse Research across 101 enterprises, 71% of respondents say a quarter or fewer of their deployed “agents” are true multi-step orchestrated workflows rather than single-prompt chatbot wrappers. Only 10% report having crossed the halfway mark.

That mismatch is happening while teams are consolidating fast onto major model-provider platforms. Anthropic’s Claude is the primary orchestration platform for 40% of enterprises, more than double the next platform, with Microsoft at 18% and OpenAI at 13%. Put differently, companies are choosing the model gravity they want, building orchestration around it, and then discovering that most of what’s actually deployed does not yet match the reliability and control they say they need.

So what’s driving the consolidation? The survey answers it directly: enterprises pick the orchestration environment closest to the frontier model they have standardized on. “Model gravity” is the single largest factor influencing platform choice, while flexibility across models and tools (17%) and ease of development (17%) are a close second tier. Security and permissions (14%) and total cost of ownership (11%) round out pragmatic considerations. Performance, like latency and memory, sits last at 4%. In plain English: teams are optimizing first for what they want to build on, not for how they’d like the plumbing to feel under load.

And the “build on” bet is landing on platforms, not open frameworks. Major model providers together account for roughly 80% of deployments, with Anthropic, Microsoft, OpenAI, Google, and Amazon making up 81 of 101 responses. Meanwhile, open frameworks like LangChain/LangGraph, and custom in-house builds that dominate technical conversations, land in single digits here. Even satisfaction scores nudge the interpretation: respondents rate their orchestration platforms 3.94 out of 5 overall, and “ease of implementation” is the weakest score at 3.85. Also crucial: 96% plan to change their orchestration approach within the year. That’s not exactly “we’re done.” It’s closer to “this works, but we’re still hunting.”

Where the ambition really shows strain is in how enterprises define success. The top optimization target is reliable multi-step execution. Task completion reliability is 32% and multi-step workflow management is 28%, for 59% of responses combined. Developer productivity (17%) matters but is secondary, and end-user experience (9%) trails. This is why the chatbot wrapper gap is so pointed. If your definition of “agent” is dependable multi-step work, then shipping mostly single-prompt wrappers is not just a technical nit. It is a product and process mismatch that can cascade into slower operations, brittle workflows, and missed ROI.

The architecture and governance implications are equally revealing. By the end of 2026, 51% expect a hybrid control plane: provider-native capabilities plus external orchestration. Only 6% expect to hand control to a provider-managed service. The fear behind that choice is explicit in the survey: vendor lock-in is the risk they fear most if control lives inside a model provider (35%). That’s the board-level version of “don’t paint yourself into a corner.” It also matches the platform choice story: firms want the frontier model benefits, but they do not want to surrender long-term operational leverage.

Control and cost are where things get tense fast. Investment follows where teams feel they must build first: agent workflow tooling leads spend (34%), with security and permissions enforcement behind it (25%). But fiscal control lags throughout. More than a quarter (27%) have no real-time way to stop a runaway agent before the bill arrives. That is not just an engineering issue, it is a finance issue. Token burn is one of the few AI costs that can scale instantly with usage and prompts, and the survey suggests many enterprises still lack the real-time guardrails needed to manage it.

There is also a methodology footnote worth respecting if you’re making strategy decisions: this is a self-selected survey of organizations with 100+ employees (n=101), based on a single June 2026 wave. It is not a pooled multi-month probability sample, so the findings should be read cross-sectionally, not as month-to-month trend claims. Still, the internal logic is hard to ignore: enterprises are choosing orchestration platforms quickly, expecting hybrid control, optimizing for multi-step reliability, and yet reporting that most deployed “agents” are not actually doing multi-step orchestration today.

For executives and boards, the stakes are straightforward. Agent programs are moving from experiments to operations, but the operational reality, control model, and cost guardrails are still catching up to the promise. The question your peers should be asking now is not “Which platform is best?” It’s “Does our deployed agent portfolio match the multi-step standard we say we’re building, and can we stop cost blowups in real time?”