By admin 
In a striking revelation that underscores a widespread immaturity in enterprise AI adoption, new research from VentureBeat Research indicates that organizations have knowingly deployed sophisticated AI agents without establishing the fundamental control mechanisms required to manage them effectively. This proactive, yet potentially precarious, approach has left many companies scrambling to retrofit their systems and processes, with significant budget allocations now earmarked for catching up. The findings, derived from five parallel surveys conducted in June across various layers of the agentic stack, paint a clear picture of an industry grappling with the rapid pace of AI innovation outpacing robust governance. Across all five measured control layers – identity, evaluation, cost telemetry, context, and orchestration – a substantial majority of enterprises, ranging from 57% to 68%, plan to either switch vendors or introduce new solutions within the next 12 months. Even more immediate is the planned transition for roughly a third of these organizations, depending on the specific control layer, within the current quarter, signaling a pressing need to rectify existing gaps.
VentureBeat Research meticulously examined five critical control areas that enterprises must solidify before entrusting AI agents with significant operational responsibilities. These pillars of agentic governance include: identity, which dictates which agent is authorized to perform specific actions and under whose credentials; evaluation, responsible for assessing the quality and accuracy of an agent’s output; cost telemetry, designed to meticulously track the operational expenses associated with each agent; the context layer, which furnishes agents with the essential business data and definitions they need to function effectively; and orchestration, the control plane that synchronizes and manages multi-step agent workflows. Each of the five accompanying reports delves deeply into one of these crucial control dimensions, providing granular insights into the current state of enterprise preparedness.
A significant disconnect exists between the nomenclature used by enterprises and the actual capabilities of their deployed AI systems. The research highlights that a substantial 71% of organizations report that a quarter or fewer of their deployed "agents" are capable of autonomously completing multi-step tasks. Conversely, only a meager 10% of respondents indicated that true, autonomous agents constitute the majority of their AI deployments. This self-awareness among respondents is particularly potent, given that 81% hold influential positions in recommending or deciding on AI purchases within their companies. The implications of this distinction are profound: a simple, single-prompt chatbot that relies on human oversight for every output requires minimal governance. However, a true multi-step agent, capable of independent action and complex problem-solving, necessitates the full suite of controls that VentureBeat Research has identified. The fact that most enterprises struggle to accurately categorize their deployments underscores a fundamental misunderstanding of the operational requirements for advanced AI. The "Agentic Orchestration" report, in particular, details this deployment challenge, emphasizing that organizations are currently facing a problem of implementation rather than a fundamental platform limitation.
The enthusiasm for AI autonomy appears to be significantly outpacing the development of robust and trustworthy evaluation mechanisms. The findings reveal a concerning trend: two-thirds of enterprises either already permit agents to initiate code or system changes directly to production environments based solely on automated evaluation results, without any human review, or are actively working towards this capability within the next year. This level of automation is particularly alarming when contrasted with the fact that only 5% of these organizations express full confidence in the evaluations that would gate such critical decisions. Furthermore, a staggering half of all surveyed enterprises have experienced a customer-facing failure in the past year stemming from an agent that had passed internal evaluations. This stark reality emphasizes the critical need to rigorously test agent evaluations against real-world production outcomes, rather than relying solely on internal benchmarks. The "Agent Reliability & Evals" report underscores this critical "reality alignment problem," highlighting that despite the evident gaps, many organizations are proceeding with production deployments anyway.
A significant security vulnerability is emerging from the practice of allowing AI agents to share credentials. The research indicates that a substantial 69% of companies permit at least some of their agents to operate under shared credentials, meaning multiple agents utilize the same API key or service account. This practice directly correlates with an increased risk of security incidents. Organizations that allow credential sharing reported experiencing a security incident or near-miss at a rate of 63.5% (47 out of 74 companies), a stark contrast to the 40.9% rate (9 out of 22 companies) observed in organizations where each agent is assigned its own unique and scoped identity. The clear recommendation emerging from the "Agentic Security & Identity" report is the implementation of scoped identity for every agent, with a particular focus on those agents that interact with critical production systems. This move towards granular access control is paramount in mitigating the escalating risks associated with AI agent proliferation.
The high-stakes investment in AI compute infrastructure, particularly in the form of GPUs, is revealing a troubling inefficiency. More than eight out of ten enterprises that operate their own GPU clusters reported utilization rates of 50% or less. Compounding this issue, only 44% of these organizations rigorously track the actual costs and returns generated by their AI compute workloads. This suggests a widespread lack of financial discipline and strategic oversight in managing expensive AI hardware. The "AI Infrastructure & Compute" report strongly advocates for a shift in focus from simply acquiring more GPUs to optimizing the utilization and per-workload cost of the existing infrastructure. This emphasis on efficiency and cost-effectiveness is crucial for ensuring a positive return on investment in AI.
A critical gap in data governance is leading to agents providing confident yet incorrect answers, directly impacting business operations. Fifty-seven percent of enterprises have traced confident, but erroneous, agent responses within the last six months to deficiencies in their own business context. These deficiencies manifest as missing or inconsistent data, including inaccurate metrics, outdated definitions, and absent critical documents. A concerning majority of these organizations have experienced such incidents more than once. The "Context Layers / RAG" report underscores that before organizations can effectively scale their AI agents, they must first establish robust governance over the data sources from which these agents draw their answers. Prioritizing the definition and management of key metrics and business entities is a fundamental prerequisite for building reliable AI systems.
Across all five surveyed control layers, the market for solutions remains highly fluid, with no single vendor holding a dominant position. Currently, the default approach for many enterprises involves leveraging the built-in tools provided by the major AI platforms they already utilize. However, the intent to adopt, add, or replace solutions is highest in the crucial area of orchestration, where 68% of organizations plan significant changes within 12 months, and 34% within the next quarter. The direction of these strategic shifts – whether towards enhancing the capabilities of existing platform-integrated tools or adopting specialized third-party solutions – remains an open question that will likely shape the market landscape over the next four quarters. This dynamic signifies a period of intense competition and innovation as vendors strive to meet the evolving governance needs of enterprises deploying AI agents.
About This Research
VentureBeat Research, through its VB Pulse program, conducted five parallel surveys in June 2026, meticulously gathering data on the critical aspects of enterprise AI agent deployment. The surveys encompassed "Agentic Orchestration" (101 respondents), "Agent Reliability & Evals" (157 respondents), "Agentic Security & Identity" (107 respondents), "AI Infrastructure & Compute" (107 respondents), and "Context Layers / RAG" (101 respondents), resulting in a total of 573 qualified participants. All respondents were affiliated with organizations employing 100 or more individuals, ensuring a focus on enterprise-level AI adoption. While the samples are self-selected and some findings should be interpreted directionally, each individual report provides a detailed methodology. The overarching pattern supported by these independent surveys is remarkably consistent: every single survey points in the same direction, highlighting a pervasive set of challenges and a clear trajectory towards increased investment in AI governance and control. VentureBeat produces both this in-depth research and VB Transform, the premier conference where these critical findings were initially unveiled.
