对个人助理扩展路线的批评：应做集体意识网络效应

Daniel Rupawalla @danielrupawalla原推发布于 2026.10.13

在 X 上打开原推

imo personal assistants (muse, dots, etc; instinct is ahead) are approaching how to scale incorrectly. i would use a collective consciousness to make network effects stronger. if i could control the roadmap i would: (1) introduce ways to connect the community and override existing infrastructure. the easiest example for this is if i am traveling to NYC and i have a lot of friends in NYC, my agent will automatically connect with theirs and figure out if anyone has an open couch/room and suggest it to them anonymously. if they accept, they can reach out and offer. i.e. if i ask my agent to find a place it'll canvas both my friends + hotels and ping them async so they can reach out to me or something. you can set up a similar network effect with cars, apartments, houses, long-term leases, food-sharing, etc. you could challenge airbnb, uber, doordash, reservations & taskrabbit all in one product and be a completely seamless approach. noah shinn referenced a similar idea for restaurants, although i think ranking people based on priority is a little unnerving since that is an alignment issue, if you have a fully built out societal graph, you could even offer the placement to someone else. "hey you and noah both have a reservation here but it seems like he's not going to get off the list, since you know him and it's his anniversary vs. a random dinner for you, want to swap?" and that feels a lot more intuitive and interesting. if i wanted to make sure everyone chose my assistant, i would try and create features that would have network effects, like the above. (2) aim to get as much opt-in context as possible. personally, i want to give up texts, files, emails, etc so that i can have the most tailored experience possible. a social graph etc would be super helpful over time if it grows with the way i treat my relationships. this helps (1) expand. processing and understanding images here would also be helpful. for example, macro-trends of new interests, fitness, etc can all be tracked via visual context and would be interesting (if im looking more tired, and i have a personal assistant handling my groceries, it can suggest foods to get or add supplements that'll help me sleep better, etc). even more interesting that feeds into (1) is if it builds out a really good social graph, if i'm throwing a party, it'll know that one of my friends has an egg allergy and proactively tell me to avoid foods like that. (3) an interesting line to draw in overstepping to make sure its not creepy, many of the references i made in (1) and (2) if introduced now may not be well received, and might take time to flesh out properly. would love to beta test naturally, but there are a lot of interesting things that need to be rolled out quickly.