Enterprises rolled out AI agents before they had the governance controls to trust them, and now they are paying for that gap. Across five parallel surveys fielded in June 2026, VentureBeat Research found that in each of the five control layers it measured, 57 to 68% of enterprises plan to switch vendors or add new ones within 12 months, and roughly a third, depending on the layer, plan to move within the quarter. That is not a vague “we need better tooling” moment. It is a budgeting and vendor roadmap being rewritten mid-flight.

The researchers broke “trust an agent” into five control layers: identity, evaluation, cost telemetry, the context layer, and orchestration. Identity answers “which agent can do what, under whose credentials.” Evaluation answers “does the agent’s work meet a standard.” Cost telemetry answers “what does each agent cost to run.” The context layer answers “what business data and definitions does it use.” And orchestration answers “how do multi-step tasks get coordinated.” Importantly, this is not theoretical. Most deployed “agents” are still chatbots in disguise. Seventy-one percent of enterprises said a quarter or fewer of their deployed “agents” can complete multi-step work on their own; only 10% said true agents are the majority of what they run.

So why are enterprises still scrambling? Because the mismatch is happening between autonomy and the controls meant to gate it. VentureBeat Research’s Agent Reliability & Evals report points to an alarming reality: two-thirds of enterprises either already allow an agent to push a code or system change to production on automated evaluation results alone, with no human review, or are actively engineering toward that within 12 months. Only 5% fully trust the evaluations that would make that call. And half of enterprises shipped an agent that passed internal evaluations and then caused a customer-facing failure in the past year.

If you are a board member or finance lead, that combo matters. Internal benchmarks can look great while production outcomes go sideways, especially when an “agent” is making multi-step changes. VentureBeat’s practical directive is straightforward: before removing human review from any workflow, test evaluations against production outcomes rather than internal benchmarks. The point is to stop treating evals like theater and start treating them like gates you can defend.

Security is another place where the autonomy train outran the rails. The Agentic Security & Identity report found that 69% of companies let at least some of their agents share credentials, meaning multiple agents can operate under one API key or service account. But where credential sharing is allowed anywhere, organizations experienced a security incident or near-miss at a 63.5% rate (47 of 74), compared with 40.9% (nine of 22) at companies where every agent has its own scoped identity. In other words: scoping identities is not just best practice. It is a measurable difference in incident and near-miss rates, and the report’s fix starts with the agents that touch production systems.

Under the hood, many enterprises are also sitting on inefficient compute and weak cost visibility, which can make governance harder to enforce. In the AI Infrastructure & Compute report, more than eight in 10 enterprises that run their own GPUs reported utilization of 50% or less. Only 44% rigorously track what their AI compute actually costs and returns. The operational takeaway is blunt: the number worth chasing first is not more GPUs. It is utilization and per-workload cost of the ones already running.

Then there is the context layer, where agents can confidently answer from data nobody is governing. The Context Layers / RAG report found that 57% of enterprises traced a confident, wrong agent answer in the past six months to missing or inconsistent business context: wrong metrics, stale definitions, absent documents. And most saw it happen more than once. This is a governance problem disguised as an “accuracy” problem. Governing the definitions agents answer from, like metrics and entities, has to happen before scaling agents that depend on them.

Finally, this catch-up is not about a single “incumbent” platform. VentureBeat notes that no layer has an entrenched default incumbent because the defaults today are the built-in tools shipped with the big AI platforms enterprises already use. Switching intent is highest in orchestration itself, where 68% plan to adopt, add, or replace platforms within 12 months and 34% within the quarter. VentureBeat’s surveys did not ask which direction that money moves, toward built-in tools or toward specialists challenging them, which means the next four quarters become a live question for anyone underwriting product roadmaps.

About the research: VentureBeat Research fielded 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), for 573 qualified respondents total at organizations with 100 or more employees. Samples are self-selected, and some findings should be read directionally; each report carries its full methodology note. Still, the pattern is consistent across every survey.

If you are building, buying, or governing agentic systems, the strategic stake is clear: autonomy is moving faster than trust, and the board will eventually ask why “passed internal evals” did not translate into safe production outcomes. These surveys suggest enterprises are already answering by retrofitting controls, reallocating budgets, and planning vendor switching across multiple layers. For peers, the question is not whether governance catches up. It is whether you catch it before your next incident, cost overrun, or customer-facing failure forces the timeline for you.