Enterprises Pull Back on Letting AI Agents Ship Production Changes Alone

What changed

Fewer large organizations are willing to let AI agents push code into production based only on automated checks. In VentureBeat Intelligence’s August VB Pulse survey, 56% of respondents at organizations with at least 100 employees said they allow that today or plan to, down from 75% in July.

Why it matters

The shift puts human review back into the release path for more companies. That makes deployment slower, but it also signals that automated evaluation is not yet being treated as a complete safety net for production changes.

The most useful reading is cautious rather than dramatic: companies may still be expanding their use of agents while tightening the conditions under which those agents can act alone. The survey covered agent evaluation, production failures and human review, but it does not show why the willingness changed or whether production failure rates changed.