Companies deployed AI agents ahead of the controls needed to manage them, and they did so knowingly. That’s the central finding of five parallel surveys VentureBeat Research conducted in June that span every layer of the agent stack. Now those companies are modernizing to catch up to their own standards, and they are budgeting for it: At each of the five layers of control we measured, between 57 and 68% of companies plan to change suppliers or add new ones within 12 months, and about a third, depending on the layer, plan to move within the quarter.
VentureBeat Research measured the five controls a company must create before it can trust an agent: identity, assessment, cost telemetry, context layer, and orchestration. Identity governs which agent can do what and under what credentials. The evaluation determines whether the agent’s work is good. Cost telemetry tracks what it costs to run each agent. The context layer provides the business data and definitions that agents call upon when responding. And the orchestration control plane coordinates the work of the agents in several steps. Each of our five reports measures one of those controls.
More deployed "agents" They are chatbots that carry the label. Seventy-one percent of companies said a quarter or less of their staff deployed "agents" can complete multi-step work alone; only 10% said that real agents are the majority of what they direct. These respondents are positioned to know: 81% recommend or decide to purchase AI in their companies. A single-message chatbot with a human reading each response doesn’t need any of the controls the other four reports measure. A true multi-step agent needs them all, and most companies can’t say which one they have implemented. (Full results: Agent orchestration report.)
Autonomy is overcoming trust in the evaluations that control it. Two-thirds of companies already allow an agent to push a code or system change to production based solely on automated assessment results, without human review, or are actively designing to achieve this within 12 months. Only 5% fully trust the assessments that would make that decision, and half of companies sent an agent who passed internal assessments and then caused a customer deal failure last year. Before removing human review from any workflow, test assessments against production results rather than internal benchmarks. (Full results: Reliability report and agent evaluations.)
Companies that allow agents to share credentials are affected more frequently. Sixty-nine percent of companies allow at least some of their agents to share credentials: multiple agents operate under one API key or service account. Organizations that allow credential sharing anywhere experienced a security incident or near miss at a rate of 63.5% (47 of 74), compared to 40.9% (nine of 22) in companies where each agent has their own scoped identity. The solution is the identity of the scope for each agent, starting with those who touch the production systems. (Full results: Agentic Identity and Security Report.)
The most expensive hardware in the building runs at half capacity or less. More than eight in 10 companies running their own GPUs reported utilization of 50% or less, and only 44% rigorously track what their AI computing actually costs and returns. The number worth chasing first is not more GPUs, but rather the utilization and cost per workload of those already running. (Full results: AI Computing and Infrastructure Report.)
Agents respond with confidence based on data no one controls. Fifty-seven percent of companies attributed a misguided, unsuspecting agent’s response in the past six months to their own missing or inconsistent business context (incorrect metrics, outdated definitions, missing documents) and most saw this happen more than once. Governing the definitions from which agents respond (metrics and entities first) has to come before scaling the agents that depend on them. (Full results: Context/RAG Layers report.)
No layer has an ingrained headline: the defaults today are the built-in tools that come with the large AI platforms that companies already use. The intention to change is highest in orchestration itself, where 68% plan to adopt, add or replace platforms within 12 months and 34% within the quarter. Our surveys didn’t ask which direction the money is moving (towards the platforms’ built-in tools or to the specialists who challenge them) and that open question is the next four quarters of this market.
About this research
VentureBeat Research conducted five parallel surveys in June 2026 under its VB Pulse program: Agentic Orchestration (101 respondents), Agent Reliability & Evals (157), Agentic Security & Identity (107), AI Infrastructure & Compute (107), and Context Layers/RAG (101) – 573 qualified respondents in total, all in organizations with 100 or more employees. Samples are self-selected and some findings must be read directionally; Each report carries its complete methodological note. What the pattern supports more strongly than any percentage is direction: each survey, independently, points in the same direction. VentureBeat produces both this research and VB Transformationthe conference where these reports debuted.
