18
20 Comments

🤯Blown Away, Everyday.

The first time I started coding with ChatGPT+ and GitHub Copilot,
🤯 I was absolutely blown away.
Those tools opened up a whole new world of opportunities...

Since then, I've generated tens of thousands of lines of code, in Python, HTML, JS & CSS and shipped 3 micro SaaS products in the creative audio space, all AI-generated:

https://www.aijinglemaker.com (10,500+ users)
https://www.aishowmaker.com
https://www.aithercast.com

The velocity of a development cycle from ideation to release is unmatched, much faster than what you could achieve even with the simplest no-code tool, and with far greater flexibility.

If you'd like to explore a personal project you've been postponing for ages and finally make it happen in a short timeframe through the magic of AI-assisted coding, feel free to ping me for a free consultation.

on March 13, 2024
  1. 3

    I am super curious: In one comment you said that you're code is fully AI generated, but you clearly have a technical and entrepreneurial background. What would you say are the most important things to understand and learn in order to create the right bits of code?

    Given your background, you seem to know how to validate ideas and develop products that solve potential customer's needs, right? That would come in handy, :) This would explain the success of your tools in such short time.

    1. 1

      Hi, thanks for your questions.

      The most important thing to understand is how the frontend is interacting with the backend. Once you get the basics of Python and Javascript interacting with each other, you can develop pretty much everything. You need to have a mind(set) which "connects the dots". I'm not a React/NextJS boy, I prefer old school Vanilla JS, which is ok for small projects shipped by a solo developer. It keeps things simple in a single deployment.

      The great thing about AI-assisted coding is that it's rather fast to go from idea to product (in a few days, once the project is clear in your head), so you can test the product pretty fast with your target audience and see if you have some form of Product/Market fit.

      It's worth noting that I didn't create any mockups or prepare any specs for my own projects (but I would not advise this approach to everyone). I like to daydream my little ventures ;)

      1. 1

        Thanks for the fast answer! I just thought, 'dude you should have a channel where you talk about your process.' - but then I saw your youtube link in another post. I just subscribed and I am eager to learn more about your journey!

        1. 1

          Thanks for joining.
          I'm also starting a newsletter at https://aicodingclub.com

  2. 2

    Are you a developer? Am I the only one underwhelmed by ai coders? I've used got 3.5, GitHub copilot, gpt 4, now mostly Gemini. Don't get me wrong they are nice and I'd keep on using it, but all the people that say like no more developers needed... The times I have to look and come back with this is wrong on either a technicality, a basic misunderstanding, not connecting to other parts that are included ect... Feels like a 50/50, due it's probably at least 75/25 in reality

    How do you find them? Any specific tips to get better responses more often the first time?

    IMHO The fastest cutest thing was auto complete by GitHub copilot for reparative things like unit tests or getter/setter type stuff, more so if you don't use strict typing and have autocomplete.

    Otherwise I most commonly used it as a more specific google, like stack exchange level but get a custom solution. (And sometimes bootstrapping stuff as an onboarding doc, which is hit and miss)

    What's your workflow?

    1. 2

      Hi, that's a great question.

      First of all, I'm more a full stack prompt whisperer than a traditional developer.
      But I can tell you that ultimately 100% of the code that I ship is AI-generated.

      If I had to pick a music analogy, I would say that I understand the score but I don't write the notes.

      The key is to master the art of chatting with your AI assistant in a computational way.

      It's all about practice. The more you do it, the better you get at it.
      From a high level perspective, my workflow is based on the pillars of a typical web architecture, i.e. how the frontend interacts with the backend, back & forth.

      My prompting style integrates the concepts that I want the AI assistant to develop in code. For instance, for a frontend interaction, I could say: "fetch in POST our /get_users Flask route, then when you get a successful response, process the returned JSON to display the following data in a table styled with Tailwind CSS: user_id, email, credits. Add an EDIT button to each row, which will trigger a modal to edit the data; later we'll create the Flask route we'll need to call for the editing part."

      The AI will deliver the JS code, ready for integration.
      Then we iterate on the result.

      After a few months of activity, I now have some heuristics to achieve what I'm expecting and I have a library of building blocks which I reuse from project to project.

      For instance, I dedicated a few hours to master the Celery (Redis) logic for parallel processing of longer tasks which would otherwise throw timeout errors in a traditional sequential process.

      I've integrated CoPilot both in PyCharm and Visual Studio (these days I use VS more than PyCharm but I started on PyCharm) and I tend to use #comments in Python and //comments in JS to prompt super effective inline suggestions (instead of just waiting for the AI to come up with its own completion). The more context you have, the better it gets.

      Exciting times.

    2. 1

      Agree with this, it has helped with productivity but its nowhere near flawless. Particularly so when you're trying to create something a little more nuanced. I have had times where the answer is entirely contrasts docs when asking about AWS product capabilities.

    3. 1

      In the end, we never wanted to have developers in the first place. A developer is someone who executes a plan written by someone else because up until now we didn't have a better way to automate manual tasks like writing code.

      What we always wanted and continue to want even with AI writing the code, are engineers. Engineering is the creative thinking part that comes before writing code. The conceptualization of what you want to do. AI helps us to automate all that is not creative work, so we can focus more on the creative part. :)

      1. 2

        It's even wider than that, it's always take a higher skilled person and put tools, systems and simplification to remove or use lower skilled work, at the the end the ultimate is to have ppl not do anything and get everything just done and supplied without needing to think of any of the moving parts.

        Some don't even consider eng. as creatives but as execution like factory workers, it's the product manager at best that is the creative manager or just the marketers

        Like I get it as I said I use it, but I feel it's clunky at best as a person that is intimate with it, I'm wondering how many true outsiders are able to work with it, with what difficulties and to what end results, it's like the web/news would have you believe that all new at least small scale software is built without a developer moving code around and I wonder if I'm doing stuff wrong

  3. 1

    I’ve been meaning to ask this question and you seem like the perfect person to answer it, how do you build and ship so fast? And I’m am talking specifics, what technologies and techniques?

    1. 2

      There's no "secret sauce" actually. I "simply" have a clear view of what I'm expecting, developed into a snappy accurate conversation with my AI assistants. And the more you do it, the better you get at it. Tech-wise, I'm using Flask/Python + HTML, (Tailwind)CSS and (Vanilla)JS. No React, No NextJS, no Vue. MySQL for the DB. All deployed on Railway.app. For small projects, it's faster since I only have to deploy one codebase.

      1. 1

        I appreciate your reply, I think I just have to do it once and come up with a system.

        1. 1

          If you're looking for one on one assistance to get things started, I'm available for coaching / mentoring (I also invite you to join my newsletter: https://www.aicodingclub.com)

  4. 1

    Have you tried the IDE called Cursor? It's a fork of VS code with GPT built in. I find it absolutely amazing to work with, you can get it to directly edit your code inline so it makes iteration really fast. You can also chat with your entire codebase, specific file or urls.

    1. 1

      I had tried it at some point but wasn't totally convinced, I've found pretty nice ways to use //comments in JS and #comments in Python quite effectively in VS Code with CoPilot. And for longer code bases, I use GPT4 or Mistral at the moment. I'm looking forward to testing Devin at some point.

  5. 1

    Is GitHub Copilot trustworthy? Maybe I'm a bit paranoid, but I feel uncomfortable giving Copilot access to my entire code base to then do with it god-knows-what.

    1. 1

      I would not worry, just don't put API Keys and other secrets outside of .env and don't run CoPilot on .env

  6. 1

    Shipping 3 saas products using AI-assisted coding. But isn't it a problem, that AI-based coding is no longer the same as it started early?

    1. 1

      Hi, I don't understand your question.

Trending on Indie Hackers
I spent $0 on marketing and got 1,200 website visitors - Here's my exact playbook User Avatar 50 comments Veo 3.1 vs Sora 2: AI Video Generation in 2025 🎬🤖 User Avatar 26 comments I built eSIMKitStore — helping travelers stay online with instant QR-based eSIMs 🌍 User Avatar 20 comments Codenhack Beta — Full Access + Referral User Avatar 20 comments 🚀 Get Your Brand Featured on FaceSeek User Avatar 18 comments Day 6 - Slow days as a solo founder User Avatar 15 comments