Anish Acharya welcomes back Josh Elman and David Pawlan, creator of Assistant Benchmark, roughly a month after their first conversation about the race to build a personal AI assistant. David Pawlan hosted an agent meetup the night before with just over 100 attendees and 10 assistant builders pitching, and says people in the agent world are craving community; Josh Elman argues the space is not zero-sum and is in a Cambrian phase.
David Pawlan has now reviewed more than 147 assistants and sorts them into four categories — general, work-oriented, multiplayer and travel — with finance yet to arrive. In his group chats of AI power users, about 3,600 messages, trust is the number one fear at 22%. Josh Elman relays Shane Mac's "picnic" idea of fences and permissions, including the incident in which a personal agent blasted Shane Mac's finances into his company Slack, and argues humans want agents to check in before acting. Anish Acharya contrasts rules with laws and principles, and the group discusses embarrassment as the real trust frontier: David Pawlan cites 1997, when 60 to 90% of internet users thought putting a credit card online was preposterous, and Hark's campaign giving $1m of $30 pizzas in exchange for full Google Workspace access.
David Pawlan reports his benchmarks. In a flower-ordering test balancing cost against same-day delivery, Instinct, Kicker and Season weighed price well while others ignored it; in an LA travel test Grokbot prioritized time, Dot prioritized cost and Instinct sat in the middle, and Grokbot was the only agent that took value from the other agents' itineraries across three rounds. On memory, all but one of the top 15 agents would export a memory file, though the markdown dossiers unsettled both David Pawlan and Anish Acharya. David Pawlan says Dot benefits from ChatGPT context and only 3% of his 354 surveyed power users make it their primary agent, and that Dazzle, from Marissa Mayer's new company, is next up for testing because of its trust layer around photos.
On commerce, 35% of surveyed users do some form of shopping through Amazon and retail, and Anish Acharya predicts the long tail wins because agents have infinite patience, while Josh Elman says Shopify is empowered and aggregators compressed. David Pawlan cites Patrick Collison's tweet about rational consumers; Anish Acharya quibbles that agents will have their own irrational preferences, such as a taste for clean APIs. Anish Acharya also argues paternalism kills products, and that people may want the opposite of productivity.
Later they discuss Wabi's threads and interest groups in Tomo, Scout's swipe shopping, gossipy agents that introduce friends, entropy and serendipity, Seattle versus San Francisco, and untapped data sources — Amazon history, home hardware MCPs, an Apple Watch with Siri recap, and agents present at dinners, Seahawks games and parties. They close with predictions: more enterprise innovation, more fun agents, and David Pawlan's mother, a technophobe, using Instinct to find a breeder for a dog like the family's late pet. Anish Acharya says whoever builds trust around what we do not want to reveal, almost like a dark agent, will have something fascinating.
Tap a name to go there · tap a line to see why they are linked