4 ms·

Great work! I think the barrier to entry here is going to gradually become smaller and smaller (as models become better and we see more multi-modal models) and

by nibab 3y ago

Great work! I think the barrier to entry here is going to gradually become smaller and smaller (as models become better and we see more multi-modal models) and the big differentiator here is that you will have lots of open source developers building functionality that is relevant to them (vs keeping this closed-source like Vercel). I am curious how much of the functionality in this area is "write code that does scaffolding and prompt engineering" vs "finetuning and model-level improvements".

muratsu 3y ago
100%!! You need one model to understand the initial prompt and figure out a development plan. Once the plan is created, the subtasks can be delegated to specialized smaller models (I think OSS will play a big role in this!).
- nibab 3y ago
  have you explored using something more lightweight than react ? for example, wouldn't it be easier to generate vanilla html with tailwindcss ?
toddmorey 3y ago
I think this kills the CMS market—-companies like Contentful & Sanity that charge $400/mo+ just for a team to be able to edit their own website. There’s been a lot of money made (and web stack complexity added) because of how much domain-specific skill is required to created and edit html content. The only missing piece is direct framework integration (Next, Astro, etc) but at the rate of development I've seen for these projects we'll have that by next week. Then you have an intuitive site builder that operates directly on the page code just like another dev on the team. One thing Vercel got really right is being able to click on any element in the page for it to become the context of your prompt. That way, you don't have to type "change the top icon inside the third pricing box to red". Also you'll get faster responses because you don't have to feed the AI the entire page / component each time.
- muratsu 3y ago
  Yeah - this space is moving really fast. Codesandbox (https://codesandbox.io/blog/meet-boxy-ai-coding-assistant https://codesandbox.io/blog/meet-boxy-ai-coding-assistant) released the click on any element functionality back in May. Funny because my original idea was to build a browser extension to modify existing pages with "click on any element and prompt" type of workflow. That approach also has certain limitations/gotchas too. eg if you click on a list element and ask it to push it down the list, it won't work. It'll fail as expected but it may be counterintuitive for someone playing with it for the first time. Having said that, I still believe that there's value in this UX and will most likely implement it soon.
- crooked-v 3y ago
  There's one huge not-immediately-obvious element there: Vercel bandwidth is expensive. My company switched our CMS content over to Prismic in part because serving all media content directly from them was significantly cheaper than doing so through Vercel, even at a fairly low volume.
  - BoorishBears 3y ago
    Vercel/Guillermo Rauch are really onto something It's established that cloud is paying a premium for not having to hire: but they've gone a step further and started testing the limits of how much you can overcharge on top of cloud pricing for good DX Clerk.dev (another investment of his often pitched along side Vercel) follows a similar model, and almost without fail if I see a Vercel sample pushing a SaaS, Rauch is usually listed as an investor
    - re-thc 3y ago
      > Clerk.dev (another investment of his often pitched along side Vercel) follows a similar model Not when there’s auth0 / Okta charging 5x more. Completely different.
      - BoorishBears 3y ago

by nibab 3y ago

muratsu 3y ago

100%!! You need one model to understand the initial prompt and figure out a development plan. Once the plan is created, the subtasks can be delegated to specialized smaller models (I think OSS will play a big role in this!).

nibab 3y ago
have you explored using something more lightweight than react ? for example, wouldn't it be easier to generate vanilla html with tailwindcss ?

toddmorey 3y ago

I think this kills the CMS market—-companies like Contentful & Sanity that charge $400/mo+ just for a team to be able to edit their own website. There’s been a lot of money made (and web stack complexity added) because of how much domain-specific skill is required to created and edit html content. The only missing piece is direct framework integration (Next, Astro, etc) but at the rate of development I've seen for these projects we'll have that by next week. Then you have an intuitive site builder that operates directly on the page code just like another dev on the team. One thing Vercel got really right is being able to click on any element in the page for it to become the context of your prompt. That way, you don't have to type "change the top icon inside the third pricing box to red". Also you'll get faster responses because you don't have to feed the AI the entire page / component each time.

muratsu 3y ago
Yeah - this space is moving really fast. Codesandbox (https://codesandbox.io/blog/meet-boxy-ai-coding-assistant https://codesandbox.io/blog/meet-boxy-ai-coding-assistant) released the click on any element functionality back in May. Funny because my original idea was to build a browser extension to modify existing pages with "click on any element and prompt" type of workflow. That approach also has certain limitations/gotchas too. eg if you click on a list element and ask it to push it down the list, it won't work. It'll fail as expected but it may be counterintuitive for someone playing with it for the first time. Having said that, I still believe that there's value in this UX and will most likely implement it soon.
crooked-v 3y ago
There's one huge not-immediately-obvious element there: Vercel bandwidth is expensive. My company switched our CMS content over to Prismic in part because serving all media content directly from them was significantly cheaper than doing so through Vercel, even at a fairly low volume.
- BoorishBears 3y ago
  Vercel/Guillermo Rauch are really onto something It's established that cloud is paying a premium for not having to hire: but they've gone a step further and started testing the limits of how much you can overcharge on top of cloud pricing for good DX Clerk.dev (another investment of his often pitched along side Vercel) follows a similar model, and almost without fail if I see a Vercel sample pushing a SaaS, Rauch is usually listed as an investor
  - re-thc 3y ago
    > Clerk.dev (another investment of his often pitched along side Vercel) follows a similar model Not when there’s auth0 / Okta charging 5x more. Completely different.
    - BoorishBears 3y ago