The agent evaluation gap: Enterprise AI organizations have a reality-alignment problem, not a coverage problem — and most are shipping to production anyway

In a surprising trend, enterprises are increasingly granting AI agents more autonomy despite concerns about evaluation alignment. Of 157 surveyed organizations, half have deployed agents that failed customers in production after passing internal checks. The biggest issue isn't coverage but a disconnect between evaluation and real-world performance. This lack of trust is leading two-thirds to push for more frequent and direct production deployments, highlighting a growing risk in the rapid integration of AI technologies.