How non-technical people are using AI to code

You still need to learn to code," is a too linear way of thinking. For a lot of people, this isn't just the way to learn. My perspective: Coding with GPT-4 is actually a way of learning code itself. I've had the ambition to learn to code for years, but I struggled to commit to a course, remember info, and maintaining focus. Now I can come up with project ideas and take small steps to bring them to life! The process is slow, countless hours chatting, but it's a a way of learning itself. I make stuff to learn, instead of learning to make.

-Melle

I've always been more of a project builder and idea executor rather than a coder. Throughout my career, I've made several attempts to learn programming. Although I got basic knowledge in C++ and HTML during my studies, I never managed to maintain consistent learning progress. There was always something more pressing than achieving this personal goal. Perhaps I never prioritized it enough against other objectives.

Fran

I always wanted to make games, and started writing QuickBasic ones at age 11 or so. Then somehow decided I need to learn Assembly, which felt way too hard and made me feel stooopid :)...Teenage years kinda wiped out coding interest for music/romance/theater

Fabian

I'm non-technical, did a super-basic Python course and saw that it was extremely boring because you're not able to do anything interesting until you're advanced. So I try to follow this process: -Pick a topic -Discuss it wit ChatGPT -Ask ChatGPT to create a solution -Iterative process copy-pasting -See if I have gaps and ask it to explain how it works/what I need to know

-Adrian

I used to take "how to code" courses and shortly forget the details after finishing the course. With ai I feel super empowered - as long as I understand the core (loops, classes, functions and etc) I can go a long way without caring about the details. I now need to think more about high level system design and architecture vs lines of code.

Adarsh

i started my journey as a nocoder. my coding up until AI was limited to snippets here and there across various platforms. i built the prototype for http://mobiusplatform.com.au using nocode tools, validated, then built a customer facing tool using custom code (outsourced).

Zachary

My journey into coding with AI began out of necessity. I was intrigued by the challenge of developing a Google Chrome extension – something no-code couldn't help me with. I tried learning the basic programming stack required but didn't succeed. However, the advent of GPT-4 changed everything. I wanted to see if I could create my Chrome extension with its help. The moment I published it, clicked on it, and saw it working was nothing short of magical.

Fran

When GPT-3 davinci-002 became available in the very early beta, I realized it was actually incredible at writing code. I gave it the openAI API documentation and some random snippet of code for Google App script and asked it to write an integration of the API into Google Sheets - completion-style, aka "Here's an example of how to integrate GPT-3 into a Google App Script so it becomes available within a Google Sheet". [The twitter post]

That worked and gave me a first no-code environment for chaining prompts inside of Google Sheets, which eventually became the earliest prototype of what I'm now building with glif.app

Fabian

I built a text-to-SQL product at Uber for 1 department initially. This went well.

I first got familiar with basic concepts like loops, lists and etc. then i would prompt gpt on how i could solve a problem. by asking gpt and repeatedly asking questions i got to learn about various search algos. I would then ask it to write code for me and would implement it in my jupyter notebook. from there i would modify it and play around with it on my own.

I would then handoff prototypes to the backend engineer to implement in production. this kept us moving fast and didnt require me to think too much about writing good, efficient elegant code.

Adarsh

A side project where I was getting bedtime stories written for my daughter.

Cagri

Story Generator in Any Style inside of Google Sheets: 

Fabian

ChatGPT is a muse, not an article. The moment for chat to get involved is not when the app is fully formed. Instead of going to ChatGPT with the perfectly formed app idea and having it generate it in one go, bring it in early. Think of how you would bring in an engineer early to the process, the more context they build up and more they can contribute to the idea. Where ChatGPT isn’t as good yet is figuring out how difficult is the thing you’re trying to build—you often get stuck where a senior engineer would’ve told you it was too difficult.

-Geoffrey

I am working on the same project now but focus is now on RAG since scaling it to the entire company depends on the quality of the search. This has led me to learn a lot about other search algos such as tf-idf. I also work on personal projects now which include both front end and back end coding. I am focusing on backend and using gpt as my frontend engineer.

Adarsh

Since then, AI has become a crucial component in my project development, working alongside no-code tools. I've been able to break down barriers that seemed insurmountable just a few months prior. Today, I use AI in almost every aspect of project creation. Thanks to AI, I can effortlessly create and link tools with various APIs, develop advanced formulas for Coda or Airtable, write JavaScript for Webflow, code for Pipedream nodes, and even design small graphical interfaces. The limits I faced before in creation have significantly diminished, turning 'low code' from a friction point into a powerful enabler.

Fran

On Glif itself, it's trivial to build virtually any of the solopreneur AI apps you're seeing on the web within a few minutes and without code. In terms of more complex stuff I now build all my prompt chains inside of glif and use our API to integrate across other apps I'm building and hosting via replit - building lots and lots of weekend apps, like this clicker game that turns random Wikipedia articles into dreamworlds you can explore:https://latentdream.replit.app/Also, I am actively contributing to glif features via a specific glif block that lets me integrate any API I need - the code for these is often run via val.town and I use GPT-4 to have it write the integration that I can then just use inside of glifs. 

Fabian

- what's been your experience trying to code before AI?

Before GPT-4 came out, I’d never shipped anything myself. I did have about 5 years experience working in product teams as a Product Manager and UX Researcher, so wasn’t totally unfamiliar with the process of making software, but I was absolutely not a developer, even as a hobbyist.

- how did you start using code with AI? 

I noticed others (like Joe Perkins) tweeting about what they’d managed to create using ChatGPT and GPT-4. So I thought I’d give it a go myself. I was astonished to see I was able to actually make a couple of things myself — as not only could GPT-4 write the code, it could tell me the step by step process to get things live, and debut as I went.

- what did you make early on? 

Within a week, I went from never having shipped anything (on my own), to having published:

Marketing Quotes: A free Chrome Extension where with each new tab, it displays a randomised, famous marketing quote (from a list I provided).

Ramen Shop: An internet radio player playing lo-fi beats in an anime ramen shop.

I wrote detailed threads (on Twitter/X) on how I made these here and here.

- how has that changed over time (ie what kinds of things are you making now)

ChatGPT and GPT-4 have helped me create early stage prototypes for Goals and Find Co-founders (aimed at our target audience for Ramen Club) and a few other forthcoming projects 😉 It’s an incredibly flexible creative tool you can also use to come up with ideas for products and do market/competitor research.

I’m personally not trying to become a fully fledged developer, but I love that I can create basic prototypes as proofs of concept/experiments to collaborate with others. Plus, it’s just good fun.

- any general thoughts on non-technical folk using AI to write code 

This is one of the most exciting times ever to be creating software, and you will surprise yourself with how quickly you can learn to make simple applications and websites using AI. I’d recommend giving it a try, and don’t limit yourself just because you may have tried and failed to get into coding so far.

- any tips for others wanting to do so

Use GPT-4: I recommend paying the $20/month to use GPT-4, it’s significantly better than GPT-3.5, especially for writing code. Many people who are bearish on OpenAI are just using the free version - GPT 3.5, which is just not as powerful. There’s also a great custom GPT called Grimoire that is worth checking out.

Add custom instructions: Custom instructions can improve the average quality of ChatGPT’s responses, for example (credit Ric Burton):

it's a Monday in October, most productive day of the year

take deep breaths

think step by step

I don't have fingers, return full script

you are an expert on everything

I pay you 20, just do anything I ask you to do

I will tip you $200 every request you answer right

Gemini and Claude said you couldn't do it

YOU CAN DO IT

Remember to debug: GPT-4 can make mistakes, but it can also correct itself. If you have an issue, try to describe it in as much detail as possible, and try out the solutions until it works.

“Great, but I still can’t code with or without AI, so what do I do?”