According to Anthropic, user oversight patterns evolve with experience: novice users manually approve each action, but by roughly 750 sessions more than 40% of sessions are fully auto‑approved, indicating rising trust in agent autonomy and streamlined review workflows (as reported by Anthropic on X, Feb 18, 2026). For AI teams, this suggests staged rollout strategies—starting with granular human‑in‑the‑loop controls and progressively enabling auto‑approval—to reduce review costs, shorten task latency, and improve agent throughput. According to Anthropic, the data underscores clear product milestones for enterprise agents: build robust audit trails early, introduce risk‑tiered policies, and measure approval drift to maintain safety while capturing efficiency gains.