85% of companies burned by an AI mistake are racing to cut the humans who might catch the next one

VentureBeat Intelligence survey data shows rising enterprise trust in automated AI evaluations, even though 49% of respondents said an AI agent or LLM feature passed testing and later caused a customer-visible problem. Companies that experienced these failures were far less likely to fully trust automated checks, but many are still pushing toward reduced human involvement in deployment decisions.

The findings point to a widening gap between confidence in AI evals and their proven ability to prevent real-world failures. As enterprises scale agent deployments, demand may grow for continuous monitoring and mitigation tools that catch issues after release rather than relying on pre-launch tests alone.