Experience with agentic AI platforms shows a relaxing of human-in-the-loop. This
By Allie K. Miller · February 28, 2026 · Curated by George's Blog
Experience with agentic AI platforms shows a relaxing of human-in-the-loop.
This could be for a few reasons.
1) The user actually gets a better sense at what AI can and cannot do and are more willing to let AI run with a full task. This is what most people assume.
2) The user gets lazier and lazier, falling into a lull of trust and the 50% auto-approve is just a reflection of giving up/overreliance. This is what most pessimists assume.
3) The user is so inspired over 1000 sessions on all the insane things AI can do that they are forced to reallocate their energy toward the "mega uses" and delegate more auto-approval to AI to maintain their productivity on the "basic uses". This is what no one is talking about.
Helpful graph and stats from Anthropic for business leaders and Claude Code lovers - it takes about 1,000 Claude Code sessions (3 months at 10 per day, 1 month at 30 per day) for users to turn on auto-approve for 50% of their conversations.