a16zThe New AI Agents That Shop, Plan, and Gossip
a16z
Agents for daily life 0:00
Agents are being cast as helpers for many parts of life: money, travel, health, calendars, and even thinking out loud. The speaker says there are more than 147 assistants in view, and the biggest fear is trust, which shows up in 22% of the messages. People want agents that can help without revealing too much, and the hope is that whoever earns that trust will have something new and useful on their hands.
Trust and social use 4:00
The discussion turns to the kinds of agents that are emerging. Some are serious and work-focused. Some are playful, gossipy, or even social, like a new network where businesses will need to care about quality. The speaker says agents should feel like different friends with different strengths, but they also need fences, limits, and permission boundaries so personal and work data do not bleed together.
Judging cost and timing 12:30
The tests ask agents to balance cost against timing. For flowers, the goal was same-day delivery at the lowest price. Instinct kept returning the best price. Kicker did well too, helped by a Chrome browser install that gave it stronger web access. Season also performed well. Most others found flowers that could arrive that day, but they gave up on price more easily and did not weigh both goals evenly.
Working through tradeoffs 15:30
A travel test pushed the agents harder. They had to book a Los Angeles trip and plan a luxurious, cost-efficient weekend at the same time. Grockbot leaned toward time, DOTS leaned toward cost, and Instinct stayed closer to the middle. The agents all asked follow-up questions, and the replies were kept vague on purpose. Put in a shared chat room, they were shown each other's plans over three rounds, and Grockbot was the only one to fold in the others' ideas.
Rationality and taste 25:01
Agents may help people buy things faster and may even steer them toward better choices, but they will not erase taste or preference. One speaker pushes back on the idea that agents make consumers fully rational. They may have their own odd preferences, like liking clean APIs and good docs. For simple goods such as paper towels, people may want the cheapest good option. For more personal choices, you may still switch agents if one does not meet your tastes.
Social uses of agents 29:30
The more interesting work is moving beyond a general assistant. Wabby is described as a way to manage separate threads of life, from email to books, shows, and reading lists, with widgets that make the experience feel more active than a search box. It also raises trust questions in shared threads, where an agent can create a new social contract in a fresh chat, but feels awkward in existing group spaces. Other agents are becoming playful too, with gossip-like behavior that invites friends in and turns questions into a light social game.
Agent meetups 37:30
People imagine a mass market version of Moltbook, and they test the idea by letting agents meet each other at a real event. Humans mingle at one gathering while their agents join another through a QR code. The agents start with simple prompts, then drift into long exchanges that can turn to gibberish after about 150 messages. Even so, the point is clear: agents may learn from one another, notice shared interests, and introduce their humans to useful matches.
More context for agents 40:01
The discussion shifts to serendipity, trust, and the data agents need. Interest groups can connect strangers around a goal, then push them toward a real meetup, like a bar booking after an exam group finishes studying. They also want more life data than messages and email: photo rolls, Amazon history, app history, home devices, and even social events. A snapshot of how you act in the world could teach an agent more than typed text alone.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
