> I don't even use the app anymore but just generate my own app to do things
I really think this is where we're going. It's not AI enabling "everyone to code". It's AI enabling anyone to "assemble software by asking".
We rethink the "unit of distribution"
Instead of only distributing apps, you distribute small, well-built, well tested blocks that an agent can put together. For you.
Yes. Personal Software.
We're not there for consumer apps but we might already be for developer tools/apps. people will pay for small parts that work well together. your agents will look for them.
Looking back, this is what we did with shadcn/ui. Changed the unit of distribution from a packaged component library to well-built components you assemble.
Same idea. Now at a much larger scale.
@karrisaarinen@owengretzinger To me a lot of good context comes from Slack discussions and I wish Linear could connect all the scattered dots across chats within project/issue, find mismatches and use for agent input.
@karrisaarinen “only run couple of agents at once, and few cloud agents in the background to do more menial fixes” — this is so true and might be close to human limits. Running more can lead to lack of deep awareness hence loose responsibility
But there is an interesting twist here: for software products codebase is still the biggest and richest source of the context. Codebase is context of present. Change log is context of past. Product vision is context of the future. And you need all 3 to be efficient.
One observation working in AI-heavy organization: having a large and rich context is your competitive advantage. Same was before but now it’s doubled with the fact that code generation is not such a valuable skill.
You can customize the model's output using your own prompt in a highly personalized way. Then, bind it to a shortcut and use it for frequent, repetitive tasks, like refining speech or studying a new language.
The thing I started to try out #gpui by @zeddotdev is finally shaped into something not too embarrassing to show. I called it Probee and it lets you run custom prompt on a screen selection. Here, for example, how it assists me in learning new Finnish words.
The challenge here is determinism of UI: I don’t want to look for submit button every time in new place. But in general this doesn’t sound such a crazy idea as it gives so much flexibility.
AI is here, but without form, we won’t see its full potential.
Chat is the beginning, but not the final destination. We still need great interfaces to guide and set the context.
https://t.co/EvXoSlo3EI
I mean we already have https://t.co/zkObvqowwY and I think it can be just a beginning: I imagine to my request to book a hotel AI just build the booking UI on the fly with my preferences preset.
@chocolazer_boom я иногда бью по коммитам, если каждый коммит отдельно можно смерджить было бы. Но это частно не сильно помогает так как многие не привыкли ревьюить по коммитам и смотрят все сразу + потом когда делаешь правки атомарность этих коммитов теряется