Enterprise AI agent orchestration — the coordination of multi-step workflows that chain model calls, tool use, and autonomous decision-making — is rapidly consolidating around model-provider platforms, with Anthropic's Claude leading by a wide margin, according to a VentureBeat Pulse Research survey of 101 enterprises. Respondents said they are choosing orchestration platforms largely because of the strength and pull of the underlying model, and they are judging success primarily by reliable multi-step execution.

The reality of enterprise deployments, however, remains far behind that ambition. Most deployed "agents" are still chatbot wrappers rather than true orchestrated workflows. Enterprises also expect the agent control plane to remain deliberately hybrid in order to avoid vendor lock-in, while real-time budget control over token consumption remains uncommon.

The VentureBeat Pulse Research wave examined enterprise agent orchestration across several dimensions: which platforms organizations use, what drives platform selection, what success metrics they prioritize, how they expect agent control to be structured, how orchestrated their deployed "agents" actually are, and how tightly they control the cost of running them.

The central finding is a significant gap between orchestration ambition and orchestration reality. Enterprises are consolidating quickly around major model platforms. Anthropic's Claude is the primary platform for 40% of respondents, more than twice the share of any rival, followed by Microsoft at 18% and OpenAI at 13%. Platform selection is driven by "model gravity" — native alignment with a state-of-the-art base model — cited by 21% of respondents. Success is measured by reliable multi-step execution, with task completion reliability at 32% and multi-step workflow management at 28%.

Yet when asked to assess their portfolios directly, 71% of enterprises said a quarter or fewer of their deployed "agents" are true multi-step orchestrated workflows rather than single-prompt chatbot wrappers. Only 10% said more than half of their deployed agents have crossed that threshold. In effect, the orchestration layer is being built ahead of the orchestrated portfolio it is intended to manage.

That gap is influencing enterprise architecture. By the end of 2026, 51% of respondents expect a hybrid control plane combining provider-native capabilities with external orchestration. Only 6% expect to hand control fully to a provider-managed service. Vendor lock-in, cited by 35%, is the risk enterprises fear most if agent control resides inside a model provider — a concern that echoes well-documented lock-in patterns from earlier enterprise technology adoption cycles, including cloud infrastructure and relational databases, where migration costs rose sharply once workloads were deeply integrated with a single vendor. Investment is following the build-out: agent workflow tooling leads expected spending growth at 34%, followed by security and permissions enforcement at 25%. Fiscal controls remain behind the curve, with 27% of enterprises saying they have no real-time way to stop a runaway agent before the bill arrives.

Methodology

VentureBeat conducted the survey as part of its ongoing Pulse Research series, with this instrument focused on enterprise agent orchestration. Responses were filtered to organizations with 100 or more employees, producing a sample of 101 respondents from a single June 2026 wave. Because this was one wave rather than a pooled multi-month sample, the report is presented cross-sectionally and does not infer month-over-month trends.

The sample was distributed across enterprise size bands. Organizations with 100–499 employees, 2,500–9,999 employees, and 50,000 or more employees each represented 21% of the sample. Organizations with 10,000–49,999 employees and 500–2,499 employees each represented 19%.

The respondent base was senior and buyer-relevant. Product and program managers accounted for 15%, CIO/CTO/CISO roles for 13%, and consultants and advisors for 13%, alongside data, AI, and engineering directors and vice presidents. "Other" functions accounted for 18%. On purchasing authority, 81% of respondents were recommenders, influencers, or final decision-makers for AI solutions, including 66% who were recommenders or influencers and 15% who were final decision-makers. Technology/Software was the largest industry represented at 44%, followed by Financial Services at 17% and Healthcare/Life Sciences at 8%.

VentureBeat said the 101-respondent sample is large enough to read directionally with reasonable confidence, while noting that it remains self-selected and is not a probability sample.

Finding 1: Orchestration runs on model-provider platforms

Anthropic's Claude leads, while open frameworks remain marginal.

VentureBeat asked which agent orchestration platform enterprises primarily use today. Responses were concentrated among major model providers, especially Anthropic.

The report cautioned that the figures should be read carefully. Respondents were self-selected, and the question asked for a single primary platform. As a result, the numbers measure which platform leads each enterprise's deployment within a self-selected audience of AI-active technical decision-makers. VentureBeat said such a sample can differ substantially from spend-weighted market measures, and each VB Pulse survey has its own sample and company-size mix. The platform shares should therefore be read as a view of where this cohort has placed its primary orchestration bet today, not as market share.

Model platforms dominate the current deployment picture. Anthropic, Microsoft, OpenAI, Google, and Amazon together account for roughly 80% of deployments, or 81 of 101 respondents. Open frameworks such as LangChain/LangGraph (open-source toolkits for building LLM-powered applications that are widely used in developer communities but have not yet translated that mindshare into enterprise production share) and custom in-house builds remain in single digits. Anthropic's 40% share, more than double the next platform, aligns with the "model gravity" selection logic described in the second finding: enterprises are choosing the orchestration layer that comes with the model they want to build on. As in VentureBeat's prior agent-security wave, the tools that define the category in technical circles are not yet where enterprise deployment is concentrated. A small 3% said they are not orchestrating at all.

Respondents rated the platforms they use at 3.94 out of 5 overall, with 109 answering that rating question. "Value for money" also scored 3.94, while "ease of implementation" was the weakest score at 3.85. VentureBeat said that places orchestration near the bottom of its five-tracker satisfaction range, ahead only of evaluation tooling. A rating just below 4 out of 5, among users of whom 96% plan to change their orchestration approach within the year, suggests provisional acceptance: the platforms work well enough for current use, but not well enough to end the search for something better. VentureBeat characterized this as a layer enterprises tolerate more than they love.

Finding 2: Model gravity drives platform selection

The base model, rather than tooling alone, is deciding platform choice.

When VentureBeat asked what most influenced orchestration platform selection, the largest factor was the pull of the underlying model. Flexibility and ease of development followed closely.

The report said the leading role of model gravity explains Anthropic's platform lead on the selection side: enterprises tend to choose the orchestration environment closest to the frontier model on which they have standardized. But the next tier of responses complicates the picture. Flexibility across models and tools was cited by 17%, and ease of development also by 17%, showing that enterprises want to avoid being trapped by their platform choice. That concern foreshadows the lock-in risk described later in the report. Security and permissions, at 14%, and total cost of ownership, at 11%, rounded out a pragmatic buying logic. Performance, including latency and memory, ranked last at 4%, indicating that at this stage of adoption the binding constraints are model fit and optionality rather than raw speed.

Finding 3: The job is reliable multi-step execution

Enterprises judge orchestration by whether it completes the work.

VentureBeat asked respondents to identify their primary success metric for orchestration. Reliability and multi-step workflow management dominated, while developer- and user-facing measures trailed.

Task completion reliability, at 32%, and multi-step workflow management, at 28%, together accounted for 59% of responses, or 60 of 101 respondents. In the enterprise view, orchestration succeeds when it reliably carries a task through multiple steps to completion. Developer productivity, at 17%, matters but is secondary, the inverse of its prominence in framework discussion. End-user experience, at 9%, is a smaller concern, consistent with orchestration being treated primarily as an internal execution problem rather than a user experience problem.

This reliability-first standard makes the "chatbot trap" finding especially pointed. Enterprises define success as dependable multi-step execution, yet most deployed "agents" do not do multi-step work at all.

The trap is not evenly distributed by company size. Among smaller enterprises, 77% said a quarter or fewer of their agents perform true multi-step work, compared with 62% of larger enterprises. VentureBeat said larger enterprises are meaningfully further into genuine multi-step deployment, while the chatbot trap appears directionally to be more of a mid-market condition.

Finding 4: Consolidate, productionize, and build in-house

Three strategic moves are nearly tied for the year ahead.

VentureBeat asked what major change enterprises expect in their orchestration strategy over the next 12 months. Three moves clustered at the top and were almost evenly split.

Building in-house control was cited by 25%, standardizing on one framework by 24%, and moving agents from sandbox to production by 23%. VentureBeat said the three are statistically indistinguishable and together tell a single story: enterprises are moving from experimentation toward operational consolidation. They want fewer frameworks, more production exposure, and greater ownership of the control layer. Only 4% expect no change.

The appetite for custom in-house control planes is notable when viewed alongside the platform concentration described in Finding 1. Enterprises are standardizing on model-provider platforms while simultaneously planning to wrap them in control logic they own, a hybrid posture that the report makes explicit in Finding 7.

Finding 5: Nearly seven in 10 plan to switch, and the largest group has no shortlist

The strategic changes enterprises anticipate are tied to vendor movement. When asked whether they plan to adopt a new, additional, or replacement agent orchestration platform in the next 12 months, respondents showed more movement in this layer than in any other layer VentureBeat tracks.

When asked which platforms they are considering, the most common answer among those in motion was that they had no shortlist yet. That group represented 29% of all respondents, the largest single response after "not considering a change." Among named candidates, OpenAI led at 16%, followed by LangChain/LangGraph at 12% and Anthropic at 7%.

The report noted that independent frameworks draw roughly double their current usage footprint in forward consideration, the same pattern VentureBeat's security tracker found for specialist vendors. Read alongside the report's concentration and lock-in findings, the picture is clear: major model-platform providers hold roughly four-fifths of current primary usage, vendor lock-in has become the leading fear, 96% anticipate a strategic change, and buyers now show intent to act, with the largest bloc still undecided. VentureBeat described the most concentrated layer of the agentic stack as also, as of June, the least settled.

Finding 6: Investment flows to workflow tooling

Tooling and permissions lead spending plans, while monitoring trails.

VentureBeat asked which orchestration-related investment will grow most next year. Agent workflow tooling ranked first, followed by security and permissions enforcement.

Workflow tooling, at 34%, is the budget-side expression of the reliability and multi-step priority identified in Finding 3. Spending is directed toward the machinery that links steps together dependably. Security and permissions enforcement, at 25%, and scaling infrastructure, at 20%, follow as the investments needed to move agents from sandbox environments into production, the strategic shift described in Finding 4. Monitoring and debugging attracted a smaller 11%, while another 11% reported flat budgets.

The emphasis on tooling, permissions, and scaling over pure observability suggests that enterprises are spending to build and harden orchestration rather than simply to monitor it.

Finding 7: The control plane will be hybrid, and lock-in is the reason

Enterprises expect to divide control between providers and their own layer.

VentureBeat asked where enterprises expect the primary control plane for agents to live by the end of 2026, and what concerns them most if that control sits inside a model-provider platform. A clear majority expect a hybrid model, and vendor lock-in is the leading reason.

Hybrid control is the dominant expectation by a wide margin, cited by 51% of respondents. Only 6% expect to hand control outright to a provider-managed service. Taken together, the hybrid, custom, and externally abstracted options — every architecture that keeps control at least partly outside the provider — add up to 88%, or 89 of 101 respondents.

The reason appears directly in responses about the risks of provider-resident control. Vendor lock-in leads at 35%, or 35 of 101 respondents, ahead of security and permissioning limitations at 28% and inflexibility across models and tools at 21%. VentureBeat said the pattern echoes the prior wave's "don't trust the model to police itself" posture: enterprises will build on a provider's platform but decline to be governed entirely by it. The hybrid control plane is the architectural hedge against the lock-in they most fear.

The June preference for a hybrid control plane marks movement from an earlier snapshot. In the April–May survey, which had 145 respondents, only 34% expected a hybrid control plane, while 12% expected to hand control fully to a provider-managed service. VentureBeat said these two snapshots do not yet establish a confirmed longitudinal trend, but the direction of the discussion is unambiguous: toward keeping control.

Lock-in has also newly become the top concern. In the April–May wave, security and permissioning limitations led at 32%, while lock-in was second at 24%. By June, the two had traded places. The concern about provider platforms appears to be evolving from whether they can be secured to whether they can be replaced.

Finding 8: The chatbot trap — most "agents" are not agents yet

Enterprises acknowledge that most deployments are still chatbot wrappers.

VentureBeat asked enterprises to assess their portfolios honestly: what share of deployed "agents" are true multi-step orchestrated workflows, rather than simple single-prompt chatbot wrappers. The response was the defining finding of the survey wave.

Combining the bottom two bands, 71% of enterprises, or 72 of 101 respondents, said a quarter or fewer of their deployed "agents" are genuinely orchestrated. Only 10%, or 10 of 101 respondents, said more than half of their deployed agents had crossed that mark.

That is the gap at the center of the report. The ambitions documented in the earlier findings — model-provider platforms, reliability-first success metrics, production rollouts, and deliberate control architecture — are far ahead of deployed reality, which remains dominated by single-prompt assistants presented as agents. VentureBeat described this less as a contradiction than as a roadmap: the platforms, budgets, and strategies are being put in place precisely because the orchestrated portfolio is still thin. The open question for later survey waves is how quickly reality closes on the ambition.

Finding 9: Fiscal control is still reactive

Only a minority can stop a runaway agent before the bill arrives.

Finally, VentureBeat asked how enterprises enforce fiscal control over agent token consumption — the risk that an autonomous loop exhausts a budget before anyone intervenes. Because LLM inference is metered per token — the text fragments models process as both input and output — an agent cycling through repeated calls without limits can accumulate significant costs before any human review. Most enterprises rely on native caps or after-the-fact monitoring, while real-time programmatic control remains the exception.

More than a quarter of enterprises, 27%, said they have no real-time, programmatic way to stop an agent before a budget-breaking bill arrives; they learn about it afterward from logs. Another 32% rely entirely on the native caps and throttles built into their primary platform, a form of control that is only as strong as the provider's tooling and ties back to the lock-in concern described in Finding 7. Enterprises building custom gateways, at 23%, or using cross-model routing to arbitrage cost, at 19%, are treating token burn as an engineering problem to be controlled deterministically.

As with orchestration maturity, fiscal control is an area where operational reality trails ambition. Agents are moving toward production faster than the cost-control plane around them is being built.

The report also noted a company-size split. Roughly one in three enterprises under 2,500 employees, or 34%, exercise only reactive control of agent spending, compared with 20% of larger enterprises. VentureBeat described those figures as directional but consistent with the chatbot-trap split: the mid-market is running the least mature agents on the least instrumented budgets.

Bottom line: The orchestration layer is real, but most agents are not yet

Organizations with 100 or more employees describe an orchestration strategy that is consolidating quickly but maturing slowly. They are standardizing, at least for now, on model-provider platforms, which collectively account for roughly four-fifths of primary usage. They are choosing those platforms because of the gravity of the underlying model, and they judge success by reliable multi-step execution. Investment is moving toward workflow tooling and permissions. Strategy is centered on consolidating frameworks and moving agents into production. The expected control plane is deliberately hybrid because vendor lock-in is the risk enterprises fear most.

But that standardization is provisional. Sixty-eight percent plan to adopt a new, additional, or replacement orchestration platform within 12 months, the highest switching intent of any layer VentureBeat tracks, and the largest group of those potential movers has not yet shortlisted a candidate. Current concentration shows where enterprises are today, not necessarily where they intend to remain.

The self-assessment tempers the ambition. Seventy-one percent say a quarter or fewer of their deployed "agents" are truly orchestrated, only 10% are past the halfway mark, and more than a quarter cannot stop a runaway agent in real time. The orchestration layer — including platforms, budgets, and control architecture — is being built ahead of the orchestrated portfolio it is meant to operate.

VentureBeat said that, at 101 respondents in a single June wave, the findings should be read as a clear directional signal rather than a precise measurement. Enterprises have decided how they want to orchestrate agents well before most of their agents are doing anything that requires an orchestration layer. The questions for subsequent waves are whether deployed reality closes the gap with ambition, and, with nearly seven in 10 buyers in motion and most of them undecided, which platforms the settled stack ultimately lands on.

The findings are based on survey responses from 101 qualified enterprise respondents at organizations with 100 or more employees, drawn from a single June 2026 wave. Because this is one wave rather than a pooled multi-month sample, results are directional rather than a confirmed trend. Respondents included product and program managers, CIOs, CTOs and CISOs, consultants and advisors, and directors and vice presidents of data, AI, and engineering across Technology/Software, Financial Services, Healthcare, and other sectors.