Subscribe

How non-technical people are using

AI to code
Steps you can take, examples of others, resources, tips and a
tutorial

Ben Tossell
January 18, 2024

Hey, this is Ben with a subscriber-only issue of Ben’s Bites Pro. A

weekly newsletter covering AI trends, ideas, business breakdowns and how
companies are using it internally.

I spent many years of my life wanting to build stuff on the internet. I tried to
learn to code several times and just kept hitting blockers.

So I built stuff with what we now call ‘no-code tools’. I pushed the limits by
building things like Airbnb without code, Instagram and other things. People
asked how I was doing it, so that became my startup, Makerpad—a no-code
tutorial platform.

Those who can’t do, teach…

I sold it to Zapier in March 2021 and now, AI has entered the chat.

Anyone can use AI to generate code to build applications but there are still
blockers. So I asked several folks how AI has enabled them to use code for
real-life projects and use cases. Thank you to everyone for their
contributions!

Here is how I broke this down:

The journey from not being able to code, starting coding with AI, and
building more complex projects

A mini-interview with Charlie, who built a Chrome extension in 45 mins

with AI

3 steps on how to learn to code with AI

Examples of others who’ve built projects with AI-generated code

A potential curriculum to help you learn to code with AI

Resources to help you

How you can build an app with AI-generated code

Tips for others looking to use AI to write code

Trying to code before AI

One of the issues of starting to code is learning and maintaining those
skills in a timeframe where distractions won’t take over.

Starting 100 days of Python to drop off on day 4 because I wasn’t moving
quickly enough, I hadn’t learnt how to build a full project or other work got in
the way, is just too common. It's happened a lot.

Prioritisation of learning this new skill vs doing work I already know is

tough.

Leaning towards tools or workYows I already Zgured out takes away from
time spent learning to code. It's like being a guitar player and deciding to
learn the piano. You have a sense of music, rhythm, and melody, but you
have to start with simple tunes and scales to master the new instrument.

Sometimes the speciZc course, learning format or not even knowing which
course is best is enough to put people off.

And sometimes life just gets in the way. Having kids, going to college, a
new job, a new house, a new relationship etc.

It feels like there needed to be a bigger NEED to learn.

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
Znishing 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
Perhaps these feel familiar.

Starting to code with AI

That need still has to be there but it’s easier to pick up writing code, thanks
to AI.

Previously writing a script felt like a mammoth task which can now be done
in seconds using ChatGPT.

The problem is no longer writing code, but understanding what it does and
what to do with it.

And needs change when the way with which to get there becomes easier.

Using a no-code tool to build something still requires you to build the thing
and takes time to learn the platform, test it, tweak it, and so on.

But there comes a time when off-the-shelf tools don’t cut it or don’t make
sense.

It’s not simple to build Chrome extensions or Google App Scripts with no-
code tools. And sometimes you just need a script to do a job for you, or you
want to work with an API.

Personal projects often get so far that you want to take them further, adding
features. It sparks your curiosity about what else you could build.

Note: This is why I always pushed for learning no-code tools Zrst, you Zgure
out what it takes to build something, test and tweak it then launch it. You
learn a lot about the process and oftentimes you want to go the extra mile,
which often means, learning to code.
 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 Zrst 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

Projects at work can be a great motivator to pick up coding. You come

across a thousand problems that could ‘just be solved if I could code’. And
collaborating with developers on your team becomes much easier.

It could improve an internal process or something that beneZts end users.

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

well.
 I Zrst 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, elcient elegant code.

Adarsh

Personal projects are another great motivator. Everyone thinks they have a
$Bn app idea but starting small is much easier to stomach and you learn
the process of building and shipping something.

You could build something for your friends, partner, roommates or kids to
generate bedtime stories, come up with meal plans or organise trips.

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

daughter.

Cagri

Story Generator in Any Style inside of Google Sheets:

Fabian

And you need to think of bringing in AI at the right moment, instead of

expecting it to spit out the perfect app from one prompt.
 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 Zguring out how
dilcult is the thing you’re trying to build—you often get stuck where a
senior engineer would’ve told you it was too dilcult.

-Geoffrey

Building more complex projects

Once the lightbulb moment has happened, further iterating on what started
as a simple project or embarking on new ones becomes less daunting.

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 WebYow,
 code for Pipedream nodes, and even design small graphical interfaces.
The limits I faced before in creation have signiZcantly 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 speciZc 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

A mini-interview: How Charlie Ward built a functioning

Chrome Extension in 10 steps and 45 minutes, with no
coding experience. Using ChatGPT + Replit

- 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-Z 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 Yexible
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 Yedged 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

signiZcantly 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 Zngers, 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?”

I spoke with many others on recommendations on how to approach this

and have grouped them into themes below that I think you’ll Znd useful.

They’re not necessarily in order—you can go in wherever you think it’s best
to help you learn.

The below breaks down as:

3 steps on how to learn to code with AI

Examples of others who’ve built projects with AI-generated code

A potential curriculum to help you learn to code with AI

Resources to help you

How you can build an app with AI-generated code

Tips for others looking to use AI to write code

 Subscribe to Ben's Bites Pro to read the rest.

Become a paying subscriber of Ben's Bites Pro to get access to this

post and other subscriber-only content.

Upgrade

Already a paying subscriber? Sign In

Join the conversation

Add your comment...

Login

Login or Subscribe to participate.

Keep reading
View more

Ben's Bites
Learn about AI. Curated news and product launches, daily. Deep
dives on business use cases, weekly. Over 100,000 subscribers

Enter your email

Subscribe

© 2024 Ben's Bites.

Privacy Policy Terms of Use

How Sandwich Video uses AI in its business.
The impact of AI in a creative industry.
14 days ago

From Pixels to Possibilities: AI Vision

2 months ago

How Mistral AI, an OpenAI competitor, rocketed to $2Bn in <12 months

How Mistral AI, an OpenAI competitor, rocketed to $2Bn in <12 months
a month ago