VentureBeat's VB Pulse survey puts Microsoft at 38.6% enterprise agent orchestration adoption, OpenAI at 25.7%, and Anthropic at a first-measurable 5.7%. The model is easy to swap; the agent runtime is not. Lock-in is moving from the model layer to the control plane.

A model is easy to swap An agent runtime is not

VentureBeat dropped a quiet-but-load-bearing piece on Friday that the agent-platform vendors do not want their enterprise customers to read. The short version: the model layer is now commoditized, the orchestration layer is the new lock-in battle, and Anthropic is just starting to show up on the leaderboard.

What You Need to Know: VentureBeat's VB Pulse survey of 70 enterprise technical decision-makers found Microsoft Copilot Studio and Azure AI Studio leading agent orchestration adoption at 38.6% in February 2026, up from 35.7% in January. OpenAI held second at 25.7%, and Anthropic captured a first-measurable 5.7% share. The headline framing — "a model is easy to swap, an agent runtime is not" — is the strategic point.

Why It Matters

The agent orchestration layer is where the next 18 months of enterprise AI lock-in will happen. The model question (GPT vs Claude vs Gemini) is increasingly a workload-routing decision. The orchestration question (whose harness, whose sandbox, whose memory, whose audit log) is a 3-5 year infrastructure commitment.
Anthropic showing up at 5.7% in the orchestration layer is the number that matters, not the percentage itself. Four respondents in this cohort moved from 0% to "primary platform" for Claude Managed Agents in one month. The model momentum is spilling into infrastructure.
Security and permissions beat flexibility as the buying criterion. VB Pulse shows security/permissions at 37-39% of selection criteria, control over agent execution rising from 17.9% to 22.9%, and "flexibility across models and tools" dropping from 35.7% to 25.7%. The market is shifting from optionality to governance.

What Actually Happened

VB Pulse data: who is leading enterprise agent orchestration

VentureBeat's independent Enterprise Agentic Orchestration tracker surveys qualified technical decision-makers at enterprises. February 2026 results: Microsoft Copilot Studio and Azure AI Studio at 38.6% primary platform (up from 35.7% in January), OpenAI's Assistants and Responses API at 25.7% (up from 23.2%), and Anthropic's tool use and workflows at 5.7% — its first measurable share. Tom Findling, CEO and cofounder of Conifers, told VentureBeat: "This is the convergence moment for enterprise AI. Models and agent frameworks have matured enough together that enterprises are now shifting focus beyond model quality to the control plane around it."

The VB Pulse Foundation Models tracker, separately, shows Anthropic at the model layer rising from 23.9% in January to 28.6% in February to 56.2% in March 2026 among qualified enterprise respondents (the March number is directional because the cohort was only 16 respondents). The pattern is the same: small numbers at the start of a new market structure, accelerating fast.

Why "a model is easy to swap, an agent runtime is not" is the right frame

A model can be routed to by task. Anthropic's documentation describes a managed agent harness with secure sandboxing, built-in tools, and API-run sessions. Once a company's workflows, tool permissions, credentials, audit logs, memory, sandboxed execution, and operational monitoring live inside one provider's environment, switching providers becomes less like changing models and more like changing infrastructure. VentureBeat quotes Ev Kontsevoy, cofounder and CEO of Teleport: "Orchestration without identity only multiplies chaos. Without identity, you don't know what an agent can access, what it actually did, or how to revoke its access when it operates outside policy." Syam Nair, Chief Product Officer at NetApp, made the data side explicit: effective agent management requires built-in intelligence and a continuously updated understanding of where enterprise data lives.

The two practitioner guides worth bookmarking

VentureBeat's piece points to two long-form practitioner reads alongside the survey: (1) the VB Pulse Q1 Foundation Models and Intelligence Platforms tracker, which decomposes enterprise model preferences by task class; and (2) the Anthropic engineering post on managed agent architecture, which frames the runtime as three pieces — the model, the session/harness, and the sandbox. Both are useful if you are picking a primary agent platform in 2026 and want to understand the lock-in mechanics before signing an enterprise contract.

The Take

The headline is right and the implication is the part most enterprises will miss. Routing models by task is a 2026 best practice. Letting a single vendor own your orchestration runtime is a 2027-2029 lock-in commitment you cannot easily unwind. If you are an architect picking Copilot Studio or ChatGPT Enterprise or Claude Managed Agents as the primary platform, the question is not "which model is best" — it is "whose identity layer, whose audit log, whose permission model am I willing to standardize on for the next 3-5 years." Microsoft is winning this on distribution, OpenAI on incumbency, and Anthropic on the model + runtime alignment argument. Pick the lock-in you can defend.

Quick Summary

VentureBeat's VB Pulse data confirms: the next enterprise fight is over the agent control plane, not the model. Microsoft leads at 38.6% adoption, OpenAI at 25.7%, and Anthropic at a first-measurable 5.7%. The model is easy to swap; the agent runtime is not.

Sources:

Source: VentureBeat | mr.technology — The Master Skill Index