When you ask an AI a question, the worst case is a wrong answer you can ignore. When you let an AI act — booking a ride, placing an order, spending your money — a mistake is harder to undo. That difference in stakes is what researchers call the delegation gap.

This week, Uber disclosed that AI agents completed just four rides and ~500 Uber Eats orders during its busiest week, against 320 million total trips. DoorDash reported a few hundred agent-placed orders since launch. The infrastructure exists; the integrations are live. The bottleneck is human willingness to hand over the wallet.

Trusting an AI to act on your behalf requires three things that answering questions does not: confidence it understands your preferences (not just your request), assurance it won't overspend or make irreversible choices, and clarity on who is responsible when it goes wrong. Until those are settled, most people find it faster to just open the app themselves.