Survey finds half of enterprises shipped agents that passed internal evaluations but failed customers

A VentureBeat Pulse Research survey of 157 enterprises found that half have shipped an AI agent or LLM feature that passed internal evaluations but later caused a customer-facing failure, with a quarter reporting multiple such incidents. Only 36% reported no such failures, while 8% run no pre-deployment evaluations and 6% do not track root causes closely enough to know.

Trust in automated evaluation is low: only 5% of organizations say they fully trust it, with 29% citing misalignment with real-world outcomes as the top limitation. Other cited weaknesses include bias or inconsistency (21%), lack of explainability (18%), and data-leakage or privacy concerns (17%).

Despite this lack of trust, two-thirds of enterprises (66%) already permit fully automated, zero-human-in-the-loop deployment for low-risk agents (34%) or are actively engineering pipelines to allow it within twelve months (33%). Only 22% rule it out for the foreseeable future.

The evaluation stack remains fragmented, with provider-native tools like OpenAI’s native evals and traces (17%) and Anthropic’s Claude Console evals (13%) tied with 17% of enterprises using no dedicated agent-evaluation tooling at all. Specialist vendors such as DeepEval (12%) and Braintrust (8%) hold smaller shares, and 11% have built their own solutions.

Production monitoring practices are similarly uneven: 51% of organizations monitor only whether the agent is functioning (uptime, latency, errors), while 23% monitor whether answers are correct. Roughly three-quarters run no automated, real-time evaluation of output correctness in production.

Enterprises prioritize cost (28%) and ease of integration (27%) when selecting evaluation tools, with evaluation consistency (36%) as the primary measure of success. Planned investments over the next year skew toward production observability (leading) and human review workflows (26%), even as organizations move toward zero-human deployment pipelines.

A majority (64%) plan to adopt, add, or switch evaluation platforms within twelve months, with 31% targeting adoption within the next quarter. DeepEval leads consideration sets (20%), followed by OpenAI’s native evals (13%) and Braintrust (9%).