Enterprise AI agents are giving confident, wrong answers more than once, most often at the companies best equipped to catch it.

AI infrastructure spending accelerates, yet most enterprises can't measure its economics. Only 21% run AI at scale, while 64% plan to switch providers.

AI agents in 107 enterprises face security gaps, with over half reporting incidents. Identity and isolation controls lag behind agent autonomy.

57% of enterprises report AI agents giving wrong answers due to context issues. The industry is building solutions, but most are not yet in production.

Enterprises grant AI agents autonomy beyond evaluation trust, risking customer failures. Two-thirds move toward zero-human deployment despite trust gaps.

Anthropic's Claude leads enterprise AI orchestration, favored for model gravity and reliable execution, though most agents remain simple chatbot wrappers.

145 enterprise leaders say AI spend is outrunning governance — most can't name a single owner for cross-platform AI, and shadow spending is already costing them.

New VentureBeat survey: the production failure point isn't the model. It's the runtime — and most are building the wrong fix.