To avoid failure, launch AI agents with high human control and low agency, such as suggesting actions to an operator. As the agent proves reliable and you collect performance data, you can gradually increase its autonomy. This phased approach minimizes risk and builds user trust.

A user deployed AI to research obscure camper van deals in Europe, message vendors, and even complete a multi-hour travel agent licensing exam to save money. This showcases using AI not just for answers, but for executing complex, goal-oriented projects.

To overcome user distrust of AI agents having access to personal data, the adoption path must be gradual. The AI should first provide suggestions for the user to approve (e.g., draft emails). Only after consistently proving its reliability and allowing users to learn its boundaries can trust be established for autonomous action.

The true power of an AI travel agent lies not just in booking complex trips, but in handling disruptions. Glenn Fogel's goal is a system that predicts potential issues like weather or mechanical failures and re-arranges the entire itinerary—flights, hotels, cars—seamlessly before the traveler is even aware of the problem.

Instead of starting with simple generative AI tasks, Airbnb focused on the most difficult application: resolving urgent customer issues like lockouts. This high-stakes approach allowed them to build a robust agent that can now be applied to less critical, "up-funnel" use cases like travel planning.

Initial adoption of AI agents was driven by solving small, personal annoyances like ordering groceries, dubbed "computer errands." This low-stakes entry point helped users build familiarity and trust with the agent before graduating them to more complex, high-value professional work.

A free trial for an AI agent hosting service revealed an unexpected user behavior: spinning up powerful AI agents for specific, time-bound tasks (like a coding project or planning a trip) and then letting them self-destruct. This concept of temporary agents opens up new possibilities beyond persistent personal assistants.

While AI models excel at gathering and synthesizing information ('knowing'), they are not yet reliable at executing actions in the real world ('doing'). True agentic systems require bridging this gap by adding crucial layers of validation and human intervention to ensure tasks are performed correctly and safely.

Early agent attempts failed because their reliability was too low. Without a baseline of success ('escape velocity'), users won't try meaningful tasks, which starves the model of the crucial usage data and feedback needed for it to learn and improve.