10+ hours of podcasts, in 5 minutes.
6 quotes from 2 episodes on TBPN and Lenny's Podcast, each with a timestamped link to the source.
Anish Acharya stated that consumer AI agents face steep unit economic hurdles and hit technical plateaus without human guidance. Operating costs reach thousands of dollars per user annually, while the average American spends only $5,000 to $6,000 on online travel and commerce.
AI agents can optimize restaurant seating capacity across an entire schedule far better than traditional reservation portals.
“I think their ability to fill the restaurant in a way that's sort of globally optimal is way higher…”
Platform earnings will divide between capturing user attention and fulfilling completed transactions.
“There's going to be a portion of your revenue that is driven by attention and a portion of your revenue that's driven by fulfilling transactions.”
Anish Acharya estimated that running agents costs thousands of dollars per user each year. Covering these infrastructure expenses requires an unsustainable take rate because the average American spends $5,000 to $6,000 annually on online travel and commerce.
“I think right now our best estimates are thousands of dollars per user per year which is obviously prohibitive…”
“The average American, just so you know, spends $5,000 to $6,000 on travel plus commerce, online travel plus commerce a year. You have to have a very high take rate.”
Autonomous loops help systems reach a local maximum before stalling. When a model makes an error, Anish Acharya stated that operators diagnose the failure by identifying what knowledge they possess that the model lacks.
“The loop will help you climb to the local maxima, but then it plateaus.”
“A useful way to think about it is anytime the model's making a mistake or doing something you wouldn't do, what do you know that it doesn't know?”
How we attribute quotes. Every quote was matched against the episode transcript, so the words and the timestamp are real (we trim filler words like "um", nothing else). The name comes from our written summary of the episode. YouTube gives us no voice-by-voice transcript, so open the timestamp to hear who is talking. See a wrong name? Tell us and we fix or remove it.
10+ hours of podcasts, distilled into one 5-minute read. Free, every Sunday morning.
One email a week. Unsubscribe with one click.