Sponsored
54% OF TEAMS HAD AN AI AGENT MAKE A MISTAKE LAST QUARTER. APPROVAL GATES DIDN'T LOWER IT.
Leadership wants agents doing more on their own, but nearly half of teams can't fully trace what agents did. The State of Agentic AI: H2 2026 surveyed 1,030 people who build, buy, or influence AI systems: * THE PRESSURE IS RISING: 25.3% adopt agents to meet leadership expectations, nearly double December's 13.5%. Teams letting agents act without approval rose from 4.2% to 7.5%. * APPROVALS AREN'T THE FIX: Teams that approve every action reported mistakes at 58%, vs. 52% with no approval. * VISIBILITY IS THE GAP: Only 53% can fully trace which agent did what last week. See the full research (ungated) →