Video yükleniyor...
Video Yüklenemedi
ai agents are local first clients --- of course they are! they're long lived processes that connect to a system, sync data, and call actions that do side effects. I wrote a post about it, and recorded a video that walks through it. link to post in reply!
36,334 görüntüleme • 1 yıl önce •via X (Twitter)
9 Yorum

"Fascinating perspective! AI agents as long-lived processes make total sense. Curious to see how this evolves. Dropping a link to your post would be great!"

anotha one

Love this. Before long the UI layer will just disappear and be replaced by an agent that draws a UI in each request-response cycle (or whatever that will be called in the future) I realise this is only tangentially related, but to me the most interesting question this touches on is what the structure of the “app singleton” from the beginning should be. How much of the structure of a program can be declared (as config rather than code)? How can we make it easy for LLMs to use? How can we formally verify that it meets some spec?

@gakonst Hey Sunil. I had been working on this same concept a data layer to enable ai agents on edge compute. Can I share YU in DM ? Would be really helpful to have your comments on it.

@gakonst sure!

In this vein are humans too?

maybe lol

I am 💯 with you on this. When I joined, our app was a couple of API calls and server actions. Towards the end of the year we switched to a hand rolled sync engine that suits our needs. It comes with a couple of caveats, but overall makes it a much better DX and UX.

awesome awesome
