AI Agents Work Best After Human Decisions
Automation is most useful when it carries a person's decision forward, rather than deciding what they want or replacing an interaction they value.
Developed from a conversation between Pete Winn and Andy David
The right moment to hand work to an AI agent is often partway through a process. A traveller can compare the available flights, choose one and let an agent complete the booking. Two people can decide that they want to meet, then leave their agents to find a suitable gap in both calendars. The agent removes the administrative follow-through without taking over the choice that gave the task its purpose.
That boundary also keeps the economics of automation sensible. Flight search already filters options by departure time, layover and airline, so assessing the resulting shortlist may take only moments. Using substantial compute to automate the entire journey might save very little. The better opportunity is the tedious remainder, such as entering passenger details and completing the transaction after the traveller has selected an itinerary.
Some tasks demand an even firmer human role. An agent negotiating a marketplace sale needs to understand acceptable prices, protect private information and act in good faith on its owner’s behalf. A calendar agent that initiates every social exchange can also remove the very contact someone wanted. People should supply the intent, judgement and relationship, then let agents handle coordination and routine execution.