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

Enterprise AI organizations face an evaluation gap where agents are granted autonomy despite a lack of trust in automated gating. Half of 157 enterprises reported agents failing in production after passing internal tests. Simultaneously, a context gap affects 101 enterprises, where retrieval-augmented generation often produces incorrect answers due to inconsistent data. While provider-native retrieval is prevalent, many firms are now developing governed semantic layers and hybrid retrieval to improve real-world outcomes and trust.