Enterprise AI teams are increasingly adopting multiple AI orchestration platforms instead of relying on a single vendor. According to VB Pulse data, the median enterprise now uses three orchestration platforms simultaneously. This trend reflects more than a desire to avoid vendor lock-in: organizations remain cautious about vendor security, authorization, monitoring, and governance capabilities.
Businesses want to maintain control over how AI agents operate, which models and tools they can access, and how much they spend. Although Microsoft currently leads in several major enterprise use cases, Anthropic has a significant lead among the platforms companies are considering for future deployments. At the same time, enterprises continue to face challenges involving token consumption, agent visibility, and AI spending controls.
These findings are based on ongoing analysis of how companies deploy and use AI in real-world environments. The research examines which AI orchestration platforms organizations select, what influences purchasing decisions, which capabilities they prioritize, what they expect from AI agents, how they manage costs, and whether their systems are truly autonomous agents or simply chatbots marketed as agents.
VB Intelligence collects feedback from technology professionals working directly in the field, including software engineers, machine learning (ML) engineers, product and program managers, and vice presidents and directors responsible for data, AI, and analytics.
Enterprise AI Teams Seek Greater Visibility and Control
Across 107 companies, AI agent orchestration is becoming increasingly distributed. The survey found that most organizations use more than one orchestration tool: 85% use at least two platforms, 64% use three, and only 15% rely on a single AI orchestration platform.
Microsoft AI Foundry and Copilot Studio appear in 70% of enterprise technology stacks. OpenAI’s Agents SDK is included in 68%, while Anthropic’s Claude Platform is used by 47%. Respondents also reported using Google’s Enterprise Agent Platform, LangChain and LangGraph, Salesforce Agentforce, Amazon Bedrock, and LlamaIndex. In addition, 22% of builders operate custom, in-house AI orchestration systems alongside commercial vendor tools.
This shift toward hybrid AI infrastructure is expected to continue. More than half of respondents, or 53%, expect their primary AI control plane to be hybrid by the end of 2026. Another 14% anticipate using provider-managed services, 13% plan to build a custom in-house control plane, and 11% expect to use an independent platform that abstracts the orchestration layer from the model provider.
The AI platform market is also changing rapidly. More than two-thirds of respondents plan to change or expand their orchestration platforms within the next year. Of those, 15% expect to make a change within three months, 24% within three to six months, and 28% within six to 12 months.
Anthropic’s Claude Agent SDK is one of the most closely watched technologies. Forty-three percent of builders are considering Anthropic-based solutions. Approximately one-third are evaluating Google’s Enterprise Agent Platform, while 31% are considering custom in-house orchestration and 25% are investigating OpenAI options.
Rather than selecting one definitive AI platform winner, enterprises appear to be preparing for a future in which multiple models, orchestration platforms, and AI agents work together across a hybrid control plane.
Overall satisfaction with current AI orchestration platforms remains relatively high, with respondents assigning an average rating of 4.17 out of 5. However, satisfaction was lower for implementation simplicity, which received a 3.91 rating, and value for money, which received a 3.63 rating. These metrics will be important as enterprise AI roadmaps and orchestration strategies mature.
What Enterprises Prioritize When Buying AI Platforms
Enterprise AI purchasing decisions are shaped by several competing priorities. Flexibility ranked first, cited by 29% of respondents. Other major considerations included security and permissions at 17%, production reliability at 15%, and control over agent execution at 15%.
Model gravity—the ability to align natively with a provider’s most advanced foundation model—was considered important by only 10% of respondents. Ease of development was cited by 8%, total cost of ownership by 4%, and latency and memory performance by just 2%.
Enterprise AI spending also reflects the growing importance of visibility, governance, and operational control. Agent monitoring and debugging account for 31% of spending, while security and permission enforcement represent 30%. Workflow tools account for an additional 19%. This marks a shift from VentureBeat’s previous survey, when workflow tools represented the largest share of AI orchestration spending.
Companies are primarily optimizing their AI systems for reliable task completion, cited by 30% of respondents; multi-step workflow management, cited by 27%; developer productivity, cited by 23%; and operational stability, cited by 13%. Only 7% identified end-user experience as a top priority, suggesting that many organizations are still focused on the infrastructure and orchestration layer rather than user-facing design.
For many enterprises, an AI workflow is considered successful when it can complete a task across multiple steps. As these systems become more reliable and widely deployed, development simplicity and end-user experience may become more important priorities.
AI Agent Monitoring and Token Visibility Remain Major Challenges
The biggest concerns when selecting an AI orchestration platform involve control, security, and monitoring. Enterprise builders do not want vendors to restrict their ability to understand what AI agents are doing or which resources they are accessing.
The most frequently cited concerns were security and privilege limitations at 37%, vendor lock-in at 23%, limited visibility and observability at 22%, and inflexible support for models and tools at 16%.
Enterprises are also struggling to control AI agent token usage. One in five respondents said they still cannot stop an agent in real time when it begins consuming excessive resources or operating outside expected limits.
Organizations use several approaches to manage AI agent costs. Thirty percent rely on native platform controls, such as built-in spending limits and throttling. Another 25% have developed custom gateway infrastructure, including proxy middleware designed to prevent runaway agents.
Twenty-five percent use dynamic model routing to send demanding workloads to lower-cost models, while 21% continue to rely solely on reactive monitoring, such as reviewing logs after an incident occurs. These organizations do not yet have real-time kill switches for stopping problematic agents.
Interestingly, company size has little impact on the maturity of AI financial controls. Eighteen percent of organizations with 10,000 or more employees still manage agent spending reactively, compared with 23% of smaller businesses.
The findings show that enterprises recognize the risks associated with uncontrolled AI spending, but many have not yet implemented the governance, monitoring, and automated controls needed to manage those risks effectively.
Most Enterprises Are Not Yet Running Fully Autonomous AI Agents
Survey participants were also asked to evaluate the maturity of their AI technology stacks. Their responses suggest that enterprise “agents” are gradually moving beyond basic chatbots, but truly autonomous, multi-step systems remain uncommon.
Only 2% of respondents said that 76% to 100% of their systems are advanced and nearly autonomous. Fourteen percent reported that 51% to 75% of their systems consist of complex, multi-agent pipelines, while 47% said that 26% to 50% of their systems are genuinely orchestrated.
At the lower end of the maturity curve, 35% of respondents said that only 1% to 25% of their systems are truly orchestrated. Another 3% reported deploying chatbots exclusively.
These results are consistent with VB’s June Pulse survey. In that study, 71% of respondents said that fewer than one-quarter of their deployed “agents” could complete multi-step tasks autonomously. Only 10% said they were deploying AI agents at scale.
Enterprise organizations are clearly investing in AI control planes, orchestration tools, monitoring systems, and supporting infrastructure. However, for many companies, the widespread adoption of genuinely autonomous AI agents remains several stages away. The immediate focus is on building the visibility, security, cost controls, and governance required to make agentic AI reliable in production.
Source: venturebeat.com