kaito
@kaito
here's how and where I used AI to build v1 of SuperPrompt. I spent ~20 hours on it from idea to sharing it here. not as glamorous as some vibe coding threads but I think it's a solid look at where it shines. added some general notes at the end. start: initial spec - my father in law told me about a sick prompt he found to practice languages and it was a gamechanger. then my friend saw ghibli pics and asked what he needed to type - hm it would be cool to make a site where you can find good prompts for anything - I brain dumped my idea into Grok and told it to ask clarifying questions and poke holes - jammed with it to narrow down a v0 list of functionality - I confirmed the tech stack I wanted to use - I asked it to write a spec for Cursor. 👉 this took about 20 min total. from start to detailed spec continued below
2 replies
1 recast
2 reactions
kaito
@kaito
Naming - I wanted to do this first because my experience with previous projects is that it’s annoying to set everything up (repo, db, auth providers) with a temporary name and changing everything later. I was confident I’d finish this project so it was worth it. - Brainstormed with ai while looking for available domains. Working from both ends, what makes sense and what’s available. - Ai sucks at naming but it lets you quickly see what you don’t want 👉 this took ~30 min? Not exact cus I think I also thought about it while doing dishes or something
1 reply
0 recast
1 reaction
kaito
@kaito
building - I tried to 1 shot it by putting the whole spec into Cursor, it generated a ton of promising files but it got stuck installing dependencies lol - so I broke it into chunks the same way I would do it without AI - I used Agent mode to set up the repo, that was fastest - then Composer to add chunks of functionality at a time. landing page, prompt grid with dummy data, auth, supabase integration with real data, search etc. 👉 needed some manual rewiring but overall much faster than doing it myself. about 4 hours to make ~80% of v0 exist. It felt basically done but little did I know.
1 reply
0 recast
1 reaction
kaito
@kaito
rewriting - Composer started getting stuck so I decided to take over. Turns out the code was a mess, everything was in like 3 files. - I decided to refactor cus the code was confusing both me and Composer - tried to get Composer to refactor but it struggled, it would split components fine but didnt reliably move all the existing functionality - I decided to rewrite a bunch of the code. it was annoying cus the app worked but I knew it'd be hard to keep adding stuff. - Used Chat to get AI help while making more controlled changes. - This made me strip a few more features from v0. I originally had upvote and user submissions but the tech debt made me realize I didnt need those to ship something usable. 👉 spent ~3 hours rewriting the whole thing with sturdier code while rethinking some features.
1 reply
0 recast
1 reaction
kaito
@kaito
Finishing - Now that the code was in good shape I started the last 20% of features - As usual that took a bunch of time. handling edge cases, details of various states, errors. Also changed functionality based on how it felt. - I was also figuring out the UI as I built it so it took some time finalizing how everything fit together. Went back and forth on a few things like how to visually highlight both prompts that output text vs images on the main page - Used grok/chatgpt to jam on these, mostly for thinking out loud. it was kinda helpful when it role played as target user personas to imagine how they would use the app - Then I was done. double checked security best practices with AI, added more prompts to the db and deployed it on vercel where I hooked it up to my domain. 👉 Spent ~10 hours on this
1 reply
0 recast
1 reaction