85% of Companies Accelerate AI Deployment Abandonment After Errors

A VentureBeat Pulse survey of 108 enterprises found that companies experiencing AI system errors post-successful testing are accelerating automation rather than slowing it. 85% of these organizations are actively adopting human-free deployment models, compared to 61% among companies without such incidents. Trust in automated checks has risen: in July, 13% of respondents fully trusted such systems, up from 5% the previous month.

However, the data raises concerns. Nearly half of respondents (49%) encountered situations where AI agents or LLM-based features passed internal tests but later caused customer-visible issues. 24% reported recurring incidents. Companies that had faced errors were six times less likely to fully trust automated checks (4% vs. 24% among those without issues).

The AI evaluation tool market is rapidly evolving. Leading tools include OpenAI’s built-in solutions (18%), DeepEval by Confident AI (17%), and Braintrust (15%)—the latter showing the highest growth, from 8% in June to 15% in July. Ease of integration (39%) was cited as the top factor in tool selection, ahead of cost (23%) and accuracy (28%). Yet production monitoring remains a weak point: only 26% of companies use automated real-time checks for AI response accuracy.

Experts note that companies increasingly rely on anomaly detection and post-deployment issue monitoring rather than pre-deployment testing. This shift stems from the growing complexity of systems, where enumerating all possible failure scenarios is impractical. Meanwhile, investments in manual checks remain critical: 38% of companies that faced errors plan to increase spending on human-involved processes despite their automation push.