after i learned about cursor.com, i went to my notes and found the with all my internet ideas and scrolled through.
at almost this same time, i'd heard an ad for bland.ai on dwarkesh patel's podcast while on a run and knew exactly the idea i wanted to try out. (fn)*
i've been the emmisarial tech support person for family members over the room. you know, guy who's 31 and wears glasses; that one? it's baked into my destiny. i don't know most things, so i'll just google or youtube mostly too find the issue they're describing. more recently, i switched to asking chatgpt.
as i'm sure anyone who works in IT will tell you, getting the user to describe the problem they're having, any errors they see, learning how to screenshot, how to look at their browser console, or god-forbid clear their cache is quite challenging. plus, i really only get asked when i'm in the room which makes me wonder how often they run into issues and just kinda don't do anything about it. or google and youtube the problem while trying to interpret. it can be kind of a lossy process.
so my idea was for a tech support buddy backed by some ai-fairy dust that could just be infinitely available and helpful. and who has the knowledge of their internet at their disposal.
if i graphed the rate of my internet ideas post chatgpt, it would be up and to the right. probably hockey stick-esque. only because things are just obviously more possible and reachable for normies than ever. my brain has gotten ahold of that and is less inclined to say, "ahh, i can't do that because i can't code."
as i've found out working on this project, anyone can code now. the difference maker is whether anyone can figure out how to not get absolutely stuck in the deployment process. in like 2 days, i have the basics of the codebase generated, the apis set up and connected through an (.env file!), and referenced correctly in the environment. i had the chat function working on the local host but i couldn't get the call function to work. i was majorly stuck on this.
feeling like i had half of it at least, i started figuring out deployment and thought i could get it working along the way. i'd used netlify's frontend for a past project and firebase's database. so, i went back to what i knew. once again, major issues with serverless functions, figuring out whether i was doing that through netlify or firebase, and how it all worked together.
to get unstuck, i looked into other options and did a bit of chatting (all in cursor) to figure out how this all works. i typically get ahead of myself trying something and then figure out how to keep going in it, or learn it, along the way.
this works for my running life, it doesn't work as well for deploying a project to the internet. i really needed to learn this.
i've done a few other blogs along the way (here, here, and here) to try and process all of it. that definitely helped.
when i got stuck on netlify and firebase, i moved to vercel. screwed that up. about that time, replit agent came out. signed up for that, tried to build in one shot, and didn't make it (NGMI). so, i went back to cursor/chatgpt to think and learn about how to construct a front and a back end, what serverless functions are.
but, some major things i learned that helped break the ice for the ship to get through
- setting up a frontend and backend through netlify
- i often think problems i'm experiencing are really big and that i just need to keep showing the terminal output code or browser console screenshots to the ai and it will have this aha! moment and reveal some deep technical knowledge from back in its brain, when really a few simple errors held me back
- e.g. 1: i'd named my serverless functions in netlfiy BLANDAI_API_KEY and the code referenced them (correctly) as BLAND_API_KEY. e.g. 2: when i couldn't get the call function to work (the piece of the app with the most value, i think), cursor kept suggesting that i create a test-bland-api.js file and run a test on that and i just kept overlooking that because it would be option 6 out 10, with option 1 being like more technical. turns out, i should've just run the test because that worked and it helped me then figure out how to adapt it to my code and make the actual function work
i've now demo'd my idea with my dad, my mother-in-law, and my wife. watching and listening to all of them interact with this thing i got on the internet was a really cool feeling. hearing them interact with an ai voice model that can have a conversation with them and talk to them about tech issues (the idea!!) was another really cool feeling. and after talking to each of them, i made changes that affected its usability. right after. that was invaluable.
so, now to go figure out how to get my first 10 users and keep hacking!
here's a little demo i recorded today.
(fn)* i don't link to flex i know stuff, only because finding these things made a difference for me and they're worth sharing.