Building Apps with GPT-4: A Guide for Non-Developers

I’ve spent the past 10 years building clients’ apps, mostly as a designer or product manager.

However, I never programmed an app myself. I lacked patience and often lost motivation due to the complex process of installing libraries and handling errors that nobody talked about. 🤯

It was frustrating. Even though I had a few ideas for my own apps, the thought of spending $5-10k on a prototype seemed excessive. I had previously tried outsourcing to Fiverr developers, but that came with its own issues.

This April, I got to thinking. There’s been a lot of buzz around GPT-4, a super-smart language model which can write essays, create poetry, and generate code. I’ve been already using it myself for some small Python automatizations, excel formulas, SQL queries, etc. So a question came to my mind…

Can GPT-4 Build a Full-Scale App?

The answer is yes.

On Easter weekend, I took this challenge to build an app just by pasting the code that GPT-4 generates based on my description. The result was HelloAI and it blew my mind. In roughly four days of simply copying and pasting GPT-4 code into XCode, I had a functional app prototype. It not only met my expectations but, in my opinion, even surpassed some competitors.

I’ve shared the app with about 20 beta testers and their feedback was super positive. After dedicating three more weeks to refining the app based on their suggestions during my evenings, I was thrilled to release it on the Mac App Store!

Two months have passed since the release of HelloAI, and I'm consistently updating the app with new features and enhancing its design. With thousands of downloads, hundreds of glowing reviews, and dozens of paying subscribers, the app's success has exceeded my expectations. The most satisfying part is managing it all single-handedly, without the need for additional developer assistance.

So How Can You Do It?

Ready to make your app a reality? It's time to break down how you can do it.

Getting Familiar with App Development

If you're a total beginner with no experience in software development, I strongly suggest spending a few days understanding the basics.

Knowing my way around software development made it way easier for me. Learn the terminology, grasp the processes, and explore what makes good software.

Don't hesitate to seek help from ChatGPT. It's a superb tutor, ready to answer all your queries. Additionally, there are several other excellent free resources to kickstart your journey:

• Software Development: www.freecodecamp.org, www.codecademy.com
• Github: https://docs.github.com/en/get-started/quickstart/hello-world
• APIs: https://www.ibm.com/topics/api
• UX: https://www.uxbeginner.com/start
• MVP Development: Lean Startup by Eric Ries
• User Testing and Feedback: www.usability.gov

Gaining a Clear Picture

Before starting, ensure you have a well-defined idea of what you intend to create. Understanding exactly what you're building, your users' needs and their usage patterns is immensely beneficial.

So, before tasking GPT-4, I recommend taking a few preparatory steps:

1. Conduct Preliminary User Research: Understand what's crucial for your app and structure your MVP accordingly. Skip all non-essential features and tweaks. Here is more about Fast & Simple User Research.
2. Seek Inspiration & Reflect: There's a 99% chance that someone has already built an app like yours. Test them out, draw inspiration, and blend the best elements you find.
3. Sketch Wireframes & Diagrams: It’s important to understand the architecture of your app, what problems you’re solving on each screen and what’s the user flow. Keep it simple. One main action per screen.
4. Choose the Right Platform: Are you targeting iOS, web, or Windows? This decision can influence many aspects of your app development process, so consider your target audience's preferences and the way how they would prefer to use the app.

Asking GPT-4

The beauty of AI is its flexibility. By adhering to certain guidelines, you can craft prompts that drive more effective responses from the AI:

1. Give GPT the right context: Send it all the related code, message from the log, used programming language, target platform, and detailed description of the intended result.
2. Be specific & exact: Just like with the real developers, fuzzy and unclear assignment leads to unsatisfying results. Apt description with proper terminology will save you a lot of time.
3. Don't repeat yourself: I've seen on the internet many examples of prompts where the author repeats the same command in several different ways. It’s not needed and just muddles the prompt.
4. Sometimes, it’s best to hit the “New Chat” button: If you're stuck with a particular problem for too long, it’s better to clear the context and start fresh. The old context tends to get compressed and confuse the AI.
5. Sample output helps: If you need to generate let's say a .json file with certain data, give an AI example of how you want to format it. You'll save yourself a few tries.
6. Understand the Token Limits: Work on your app piece by piece and don’t overwhelm AI with too many lines of code. To estimate these building blocks, you can ask AI to generate a pseudocode first or “How would you approach this issue.”

What's the Catch?

Think of GPT-4 as a team of experts in your pocket. It enables quick prototyping of simple apps, even for those with basic coding skills. However, there are a few things you should be aware of.

Firstly, GPT-4's context window works best with about 250 lines of code. Bigger, complex features might be tough to handle if they can't be broken down.

Secondly, the result isn't instant. Building an app with GPT-4 requires patience, clear instructions, error correction, and thorough testing. I assume that it takes approx the same amount of time as what the real developer would need.

Thirdly, the result won't be perfect. I let my friend iOS developer evaluate the GPT-4’s output. He noted several issues and possible future problems with the code quality. It's great for prototypes and minor tweaks, but perhaps not for security-sensitive products.

Lastly, choosing popular platforms and languages will yield better results since GPT-4 was trained on public forums and GitHub. Established languages had more robust training data and therefore AI can produce better quality code.

In Conclusion

From my point of view, GPT-4 is a game-changer, slashing the barriers to entry for aspiring app developers and entrepreneurs. Even with its flaws and issues, it can produce reasonably good software for MVP purposes and testing the product-market fit.

It'll be interesting to observe how AI-driven development evolves over time and how its limitations are addressed. I'm sure there are already dozens of no-code or low-code tools in the works that will utilize the power of GPT-4 and make it even easier to bring your ideas to life.

Will there come a day when AI can perfectly mimic a senior developer even for more advanced problems, or will there always be a need for human touch? Well, I guess we'll have the answer within a few years.