For those who’ve been with me from the start, I want to give you an update, maybe a recap of the last nine weeks and what I’ve learned building 20 apps in 20 weeks with ChatGPT and a friend. I’m also excited to share a pivot I’m taking based on what I’ve learned, what I think I’ve proved, and what I want to do next.

The Journey So Far: 9 Apps in 9 Weeks

Over the last nine weeks, I’ve built and released nine apps using ChatGPT to code. Now, I’m not a coder by any stretch of the imagination. I work as a technical product manager, but coding isn’t my forte.

The goal of the 20 apps in 20 weeks challenge was to prove that anyone can, with an idea, a bit of time and effort, and a modicum of technical skills, use ChatGPT to code and release a fully functioning minimum viable product (MVP) very easily.

Why I Took on This Challenge

I’m one of those people—and I hope you are too—who think they have great ideas every now and again. But I’ve either needed to find money to hire developers or learn how to code it myself. I never learned to code, and I didn’t have the money. I think a lot of people are in that situation, but they do have great ideas.

I always dreamed of having a big development agency with a room full of developers where I could come in, show the plan of what I wanted to build, and we could just build it. Now, with ChatGPT being released over the last 18 to 20 months, I don’t need that anymore. Literally, I can build an entire product in a weekend.

I think the last nine apps have proved that it’s possible. Now, let me tell you what I’ve learned along the way.

Lessons Learned Along the Way

1. The Power of User Stories in AI Development

Traditionally, when you build software, you go through these steps: ideation, then UI design, then you start to code the frontend, then the backend, or some variation of that.

In this situation, where I just wanted to solve one problem, I’ve eliminated any design step and gone straight to code. The reason I’ve done that is because of the tools that are available.

What I found is the key step when using AI as your developer is down to user stories, and that’s it. The better you plan your user stories, the faster you can actually get something output and coded. The AI can fill in the gaps a lot on UI, especially when you’ve used tools that will code UI for you. But it’s the user stories that you give it that get the output that make it really easy.

It’s really about knowing what the output is and then understanding the stories—the steps that people take. What’s the screen they see? What do they do on each step and on the page to get them to the output that you want?

This reliance on the stories—the journey, describing the journey—is, I think, the way that development will go. I won’t need to, and people won’t need to, actually write the code that delivers that output or that function—that will be unnecessary. But they will have to be able to describe it better.

I think the new coding language will be user story creation.

2. Focusing on Needs vs. Shoulds in MVP Development

The second big lesson has been about paring down the functionality you think your MVP needs. And this, for me, has been a hard one. I’ve always sort of over-scoped projects. But when you think about, “I want my users to be able to X,” what do they need to do to deliver that?

The keyword is need. What do they need to be able to deliver that? Not what do you want it to be, but what do they need

An example would be in the Collectify app I built, which I think was app number five or six. Users needed to be able to create an account, create a listing with a name and a photo, make some settings, and do some basic organization on the page. Those are the needs, right?

When I was building, I spent time trying to organize—to give the user the ability to drag and drop. On the left in the sidebar, there’s all of their collections. And I thought, “Alright, they should be able to reorganize and drag and drop.” And the problem was, should is not a need. I spent a couple of hours messing around trying to make drag-and-drop work. Now that should wouldn’t have stopped someone from creating the collection; it wouldn’t have stopped someone from doing what they needed to do to get to the output of creating a collection of images and information that they can keep, store, and share.

That was the need.

I think that’s a good example of when you’re using AI or you’re building an MVP. Keep going back and asking, “Is this a need for this output, or is it a should? They should be able to, but if they can’t, does that stop them from getting to the endpoint?” If it doesn’t, then forget about it. That’s a nice-to-have.

It’s that kind of lesson that has been reinforced to me in this project about really redefining how you put an MVP together.

Feedback and the Two Camps

Now, the feedback I’ve received falls into two camps.

The Curious Creators

First, there’s the people who are kind of intrigued, very much like myself, have an idea but don’t know where to start. It’s great—“Oh, you’re doing this cool, but how do I even start? Where do I go? What do I do? What’s the very first step?”

The Skeptical Developers

Then there’s the second camp, primarily developers. The common themes are:

• “Why are you building 20 things in 20 weeks? Just focus on one thing.”
• “It’s low-quality code; it’s a crappy piece of software; it doesn’t do anything; it’s basically junk; you’re just taking up space.”
• And the biggest one: “That’s fine for a short thing, but AI can never code anything of big value. It can’t code a big product.”
•  

I’m going to address that last point, pivoting the whole structure of the challenge.

Pivoting the Challenge: Building 10 Features in 10 Weeks

So I’m moving from “Me, a friend, and ChatGPT are building 20 apps in 20 weeks” to “Me, a friend, and ChatGPT are building 10 features in 10 weeks.”

The Future of Development: Focusing on User Stories, Not Code

As developers—or AI developers—we need to focus on user stories, not code.

The way that the tooling is going—and it is happening incredibly fast—is that it doesn’t matter about the code that I’m writing, whether it’s secure or not, because in the background, the tooling is now at a point where it can automatically write code, then write a test for that code, then run the test, then assess its security, rewrite the code, and have a functioning piece of code.

That’s essentially what developers do. They write code, they write a test, they test it works, and then they move on. So when they write another function, they make sure that everything runs so when they deploy that function, it doesn’t break something else because it has a test to make sure it works.

The tooling is getting to a point now where it can automatically write tests that run, and then just on the fly, rewrite the code so it will continue passing the test and not break any other test.

I’m being very high-level with that, and of course, we’re still not quite there, and we’re not quite there with codebase volume—some of these codebases are massive.

But it is what is happening; it is how development is going, and no matter how many developers don’t like it. We are at the industrial age of code. You do not do it by hand. Like you don’t weave by hand anymore. You have a machine that does the hard work. You just have to know the instructions to give it.

And it is the future.

Introducing Datatruck

Now let’s talk about Datatruck.

Datatruck was the sixth or seventh app that we released. It was about creating a tool that allowed you to have a contact form on a website, then using AI to route that contact form’s contents—read it, prioritize it, understand it, add some tags, summarize, and then select from a list of emails who to forward that one contact form to. That was the essential part of it.

It was a form forwarding with a little bit of smart routing inside—that was the concept.

But we believe there’s more that we can do here, and that’s what we’re going to do.

The Vision for Datatruck

Expanding the Sources

The idea is to break down the concept into three sections:

1. Sources: Right now, we have a contact form sent by JSON. I want to add more sources:

• Email: So we can process incoming emails.
• Hosted Forms: Essentially, it’s a form you’ll create, and it gives you a URL that you can either share or embed in your website.

Enhancing the Sorting

2. Sorting: This is where the magic happens.

• Enriching Data: I want to call APIs that will add additional information to that data before it does something. For example, in a food tracking service, I could use an API to get actual calorie information from a database.
• LLMs as Functions: My goal is to use large language models as functions. For instance, formatting data without writing code every single time. The AI can handle edge cases and dynamic changes more efficiently.

Diversifying the Destinations

3. Destinations: Currently, the destination is just email.

• Tables: So you could store information directly.
• JSON: You might want to send it to another tool, using us as a formatting intermediary.

This is what we’re going to build. There’s more than 10 features in here, but this is roughly the roadmap.

The Development Process

Starting with User Stories

The first thing I’m doing is, of course, writing my user stories. Then I’m using a tool called V0. V0 is free, but you get a limited number of messages that you can send and receive.

I wrote out the user stories and used ChatGPT to help me. I talked through what the steps are, what I want someone to make, and then I said, “Write me user journeys.”

Breaking Down the Code

I found that when using AI, it’s important to make sure all of your functions, all of the little bits, all of the steps are small. While it can output safely around 200 to 250 lines of code in one go, it can be very problematic if it gets longer. So I basically told it to reconfigure everything, breaking it down into the different steps that I want.

Overcoming Challenges with ChatGPT

When integrating with ChatGPT, I faced errors and challenges. Sometimes you go around in a circle: it says, “Here’s an error,” gives you the code to fix it, but then you get another error. When that happens, I switch between different models of ChatGPT to help resolve the issue.

Progress So Far

I’ve been working at this now for maybe two or three hours, and I’ve taken it from the initial step to where it is now, with errors after making the designs—in a few hours.

There’s still a bit of work to do, but I have an entire week. I still need to bring in functions that already existed, so there’s still some work to do, but at least you can sort of see the process. You understand what we’re building.

I’m hoping that this new journey, this new stage, is interesting enough for you to follow along.

Wrapping Up

If you like this, or if you have questions, please feel free to comment, DM, message—whatever. If you want to look at the other projects I’ve built, go to businessdaddy.org, and you’ll see all of the other projects and things. Some of them are free; some of them are paid. The only reason I’ve got paid versions is because if there’s actually a cost to running it—I’m not making money. I don’t have funding; it needs to cover its cost, and the free ones do have a donation option if you want.

In making this video, I used two of the other products I built:

• SendHub: The form that could send to an endpoint, which I will probably roll into Datatruck as a part of it.
• QuickMap: To show you the app workflow.
•  

I hope you don’t feel that by pivoting, I’ve gone off this challenge and I’m not delivering. I think it’s still a valid challenge; it’s still a large amount of coding that’s happening, but there’s an added level of complexity now. It’s also proving the end step—that you can, with AI, not only build a single-page tool or a free Chrome extension but actually build something that has complexity and larger functionality.

So again, like, subscribe, share this with your friends, watch it 150,000 times. Post it everywhere you can.