Turn on any tech podcast and you'll hear that "Agents" are the next iPhone moment. The promise: an AI that books your flights, answers your emails, and manages your life. The reality for consumers? A lot of friction and very little trust.

Agents are indeed booming in enterprise (see our Deep Dive), but enterprise environments are structured. They have APIs, documentation, and defined success execution paths.

Consumer life is chaos. "Book me a table for dinner" involves knowing that your partner hates sushi on Tuesdays, that the babysitter cancels if it's raining, and that you secretly prefer the place with the cheap wine. Current LAMs (Large Action Models) struggle to navigate this implicit context.

Would you let an AI spend $2000 of your money on a vacation package without double-checking every single detail? For 95% of users, the answer is "No." If you have to verify every step the agent takes, the time saved by "automation" evaporates. Until agents achieve 99.9% reliability, they remain a novelty, not a utility.

To be truly useful, a consumer agent needs access to everything: your bank, your email, your location, your health data. In an era of data breaches, handing the "keys to the kingdom" to a black-box model hosted by a tech giant is a hard sell. We expect the adoption curve to be much flatter than predicted.