By admin 
The era of a single, monolithic orchestration platform for enterprise Artificial Intelligence (AI) initiatives is rapidly fading. In a significant shift reflecting growing maturity and a pragmatic approach to risk management, the median enterprise is now actively managing three distinct orchestration platforms concurrently. This deliberate diversification, as highlighted by VB Pulse data, is not a consequence of haphazard adoption but a strategic response to the inherent uncertainties and perceived limitations of entrusting a sole vendor with the critical task of managing complex AI deployments. While the desire to circumvent vendor lock-in and preserve operational flexibility remains a paramount concern, a deeper undercurrent of distrust in the security and permissioning capabilities of many platform providers is driving this multi-platform trend. Enterprises are increasingly asserting their autonomy, prioritizing the ability to implement their own robust governance and security protocols.
Microsoft currently holds a leading position in primary AI platform usage, with its AI Foundry/Copilot Studio appearing in a substantial 70% of enterprise tech stacks. However, the landscape is dynamic, and Anthropic is emerging as a clear frontrunner for future considerations, demonstrating a commanding lead in what enterprises are actively exploring for their next generation of AI tools. This dual dynamic – current adoption dominance and future exploration – underscores the fluid nature of the AI platform market. Despite these advancements, enterprises continue to grapple with a persistent set of challenges. Foremost among these are the complexities surrounding token usage, a critical cost driver in generative AI, and the overarching lack of granular visibility into the spending patterns of their AI agents.
These insights are derived from an ongoing, in-depth analysis by VB Intelligence, which meticulously examines the practical realities of AI deployment and utilization within enterprises. The research delves into the platforms enterprises are actively choosing, the guiding principles behind their decision-making processes, their core priorities, their evolving expectations for AI capabilities, their strategies for cost containment, and crucially, the true nature of their AI deployments – whether they are genuinely agentic in their multi-step, autonomous operations or merely sophisticated chatbots rebranded with the "agent" moniker. VB Intelligence is actively gathering feedback directly from the frontline builders and strategists within these organizations, including software and machine learning engineers, product and program managers, and vice presidents and directors of data, AI, and analytics. This direct engagement with practitioners provides an invaluable, ground-level perspective on the challenges and opportunities shaping the enterprise AI frontier.
Concerns Around Retaining Visibility and Control Drive Pluralistic Orchestration Strategies
The findings from the survey of 107 enterprises paint a clear picture: agentic orchestration has definitively become a pluralistic endeavor. The data reveals a strong aversion to single-vendor dependency, with a significant majority of enterprises actively avoiding commitment to any one platform. A striking 85% of respondents are currently utilizing two or more orchestration tools, and a substantial 64% are managing three. In stark contrast, a mere 15% of organizations remain committed to a single orchestration platform, underscoring a widespread strategic pivot towards a more diversified approach.
Within this multi-platform ecosystem, Microsoft AI Foundry/Copilot Studio has established itself as a dominant force, present in 70% of surveyed enterprise stacks. Close on its heels is OpenAI’s Agents SDK, utilized by 68% of organizations, followed by Anthropic’s Claude Platform, which finds a place in 47% of deployments. The survey also indicates a broader spectrum of adoption, with builders expressing to some extent their use of Google’s Enterprise Agent Platform, the open-source framework LangChain/LangGraph, Salesforce Agentforce, Amazon Bedrock, and LlamaIndex. Further demonstrating the drive for customization and control, 22% of builders are augmenting their vendor-provided tools with custom, in-house orchestration solutions, signaling a growing trend towards hybrid and bespoke AI infrastructure.
This trend of "hybridability" is not a transient phenomenon; it is projected to accelerate. More than half of the respondents (53%) anticipate that the primary control plane for their AI operations will adopt a hybrid architecture by the end of 2026. Looking further ahead, 14% of enterprises envision relying on provider-managed services, 13% plan to develop and maintain their own custom in-house control planes, and 11% are strategically positioning themselves to leverage external platforms that offer abstraction from specific model providers, thereby enhancing interoperability and reducing dependencies.
Complementing this shift towards hybrid control planes, a significant majority of respondents are actively planning platform changes within the next year. Over two-thirds of those surveyed indicated an intent to migrate or expand their orchestration capabilities. Specifically, 15% are looking to make changes within the next three months or sooner, 24% are targeting a three-to-six-month timeframe, and another 28% are planning for changes within the six-to-12-month window. Among the platforms drawing significant attention for future adoption, Anthropic’s Claude Agent SDK stands out, with 43% of builders actively exploring its capabilities. Roughly one-third of respondents are considering Google’s Enterprise Agent Platform, while another 31% are focused on developing custom in-house orchestration solutions. OpenAI’s evolving suite of agentic tools is also under active investigation, with 25% of enterprises exploring their options. This widespread intention to change platforms underscores the dynamic nature of the AI market and the ongoing search for optimal solutions.
Perhaps drawing lessons from the lessons learned during the formative years of cloud computing and the subsequent challenges of vendor lock-in, enterprises are consciously avoiding the designation of a single "winner" in the AI orchestration space. Instead, they are strategically building for a future characterized by interoperability, where multiple orchestration platforms, diverse AI models, and a variety of agents can seamlessly collaborate and communicate across a hybrid control plane. This forward-thinking approach prioritizes adaptability and resilience over the perceived simplicity of a singular vendor solution.
While enterprises are generally satisfied with the platforms they currently employ, rating them at an average of 4.17 out of 5 for overall satisfaction, their enthusiasm wanes when it comes to ease of implementation, scoring it at 3.91 out of 5, and value for money, which received a rating of 3.63 out of 5. These satisfaction metrics will be critical indicators to monitor as orchestration platforms mature and AI roadmaps continue to evolve, potentially leading to shifts in adoption patterns and vendor dominance.
Where Enterprises Are Strategically Allocating Their AI Investments
The contemporary logic driving enterprise purchasing decisions for AI orchestration platforms is now a nuanced amalgamation of several critical factors. Beyond the imperative of flexibility, which was cited by 29% of respondents as a primary consideration, other key determinants include security and permissions (17%), production reliability (15%), and robust control over agent execution (15%). Interestingly, a relatively small proportion, just one in ten respondents, identified "model gravity" – the inherent alignment of an orchestration platform with a state-of-the-art base model – as a significant factor in their purchasing decisions. Ease of development was named by 8% of respondents, total cost of ownership by 4%, and latency and memory performance by a mere 2%, indicating a clear prioritization of strategic control and operational robustness over raw performance metrics or cost efficiencies in the initial decision-making phases.
The allocation of spending further reflects this enterprise priority on visibility, security, and control. Builders are channeling the most significant investments into agent monitoring and debugging tools (31%) and security and permissions enforcement solutions (30%). Workflow tooling, which facilitates the creation and management of multi-step AI processes, accounts for an additional 19% of spending. This represents a notable shift from VentureBeat’s prior analysis conducted just a month earlier, where workflow tooling had outright led AI orchestration spending, highlighting a recalibration of priorities towards operational oversight and security.
Enterprises are largely optimizing their AI investments for task completion reliability (30%), efficient multi-step workflow management (27%), enhanced developer productivity (23%), and overall operational stability (13%). The end-user experience, a critical factor in many consumer-facing applications, is currently a lower priority for a significant portion of enterprise deployments, with only 7% of respondents naming it as a top concern. This suggests that many organizations are still in the foundational stages of implementing and governing AI orchestration, with a primary focus on ensuring that the underlying infrastructure and processes are robust and effective, rather than on the immediate polish of the end-user interface. The focus on multi-step workflow management implies that enterprises recognize the value of AI agents in automating complex, sequential tasks. Simplifying development and enhancing end-user experiences are likely to become more prominent concerns as these foundational platforms become more firmly established and AI capabilities mature towards broader user adoption.
The Persistent Visibility Problem: Reining in Agent Spending and Ensuring Control
The paramount concerns for builders when evaluating and selecting AI orchestration platforms revolve around maintaining stringent control and comprehensive oversight. A significant apprehension exists regarding vendor platforms that might inadvertently constrain their ability to thoroughly monitor and understand the activities of their AI agents. Key factors at the forefront of their minds include security and permissioning limitations (cited by 37% of respondents), the perennial fear of vendor lock-in (23%), a lack of adequate visibility and observability into agent operations (22%), and inflexibility in adapting to different models and tools (16%). This underscores a fundamental desire for transparency and control over the entire AI agent lifecycle.
Compounding these challenges, in these nascent stages of AI agent deployment, enterprises continue to struggle with effectively managing and controlling agent token usage. A substantial one in five organizations admits to being unable to halt the escalating spending of a runaway agent in real time, highlighting a critical gap in financial governance and operational oversight.
To mitigate these escalating costs and maintain budgetary control, builders are implementing a variety of strategies. Thirty percent of respondents rely on native platform controls, such as built-in budget caps and throttling mechanisms, to manage agent consumption. Another 25% have proactively developed custom gateway plumbing, employing proxy middleware to intercept and manage potentially runaway agents. A quarter of respondents are utilizing dynamic routing, a sophisticated approach that intelligently offloads computationally intensive tasks to lower-cost AI models. However, a significant 21% still depend solely on reactive monitoring, such as post-hoc log analysis, indicating that these enterprises lack real-time mechanisms to prevent overspending, leaving them vulnerable to unexpected cost escalations.
An intriguing finding from the survey is that, unlike in previous assessment waves, organizational size appears to have minimal impact on the maturity of fiscal control mechanisms. Eighteen percent of enterprises with 10,000 or more employees exercise only reactive control over their AI spending, a figure remarkably similar to the 23% observed among smaller organizations. This suggests that the challenges of implementing robust, real-time cost controls for AI agents are pervasive across the enterprise spectrum, irrespective of company size or existing IT infrastructure. Clearly, while the financial implications of uncontrolled agent behavior are widely recognized, many organizations have yet to fully instrument their AI stacks with the necessary tools and processes to effectively rein in escalating expenses.
The Majority of Enterprises Are Still Developing True Multi-Step Agents
When builders were polled and asked to provide an honest assessment of their current technology stacks, a consistent consensus emerged: the evolution of "agents" beyond sophisticated chatbots, albeit still wrapped in that increasingly common descriptor, is a gradual but discernible process. The data indicates that the journey towards fully autonomous, multi-step agentic behavior is still in its early to mid-stages for most organizations.
The breakdown of these self-assessments reveals a spectrum of advancement. A small but significant minority of respondents (2%) report that between 76% and 100% of their AI systems are characterized by advanced, largely autonomous operations. A more substantial segment, comprising 14% of the surveyed builders, indicates that 51% to 75% of their systems are engaged in complex, multi-agent pipelines, demonstrating a high degree of interconnectedness and automated workflow execution. Furthermore, 47% of respondents estimate that 26% to 50% of their systems represent true agentic orchestration, signifying substantial progress in developing sophisticated AI workflows.
On the other end of this spectrum, a considerable 35% of organizations report that only 1% to 25% of their systems are truly agentic in their orchestration capabilities, with the majority of their AI deployments still functioning as basic assistants. A small but notable 3% of respondents acknowledge that they are still exclusively deploying traditional chatbots, indicating that the adoption of agentic paradigms has not yet permeated all levels of their AI initiatives. This aligns closely with VentureBeat’s June Pulse survey findings, which indicated that 71% of respondents reported that a quarter or fewer of their deployed "agents" possessed the capability to autonomously complete multi-step work. Moreover, only a tenth of organizations stated they had successfully deployed agents at scale, underscoring the challenges and timeframes associated with achieving widespread agentic adoption.
While it is undeniable that enterprises are actively investing in and building the necessary control planes and underlying infrastructures to support sophisticated AI agents, the full realization of the truly agentic wave appears to be an objective that, for many, still lies on the horizon. The current focus on platform diversification, control mechanisms, and foundational workflow management suggests a strategic phase of development and consolidation before the widespread deployment of highly autonomous and complex AI agents becomes the norm.
