Building apps in real time using LLMs

Building apps in real time using LLMs

The time to go from idea to reality is shrinking.

We just launched our biggest update to Create since our initial launch on Product Hunt: ✨ real time app building

You can try it from homepage.

So what is real time app building?

It's a much faster way to get exactly what you want. Just describe how you want the app to look and behave, and AI builds your app alongside.

Start typing, builds in real time

If it's not quite right, just add more details and the AI adds them to the next version.

Add details to your spec

You can even drop photos into the spec, tell the AI to "make it look like this" and voila, out comes the equivalent code running live.

After the AI is done, you can test your changes in the interactive 'Demo' mode. Or you can add more specific edits by tapping on a part of the app and sending a command in 'Build' mode.

When ready, you can go live with one tap to have others use it.


What's apparent as you use this new way to build is that it feels faster.

In the first version of Create, you built your app through a series of back and forth commands to the builder. You'd say want you want (e.g. "make me a landing page for a SaaS company that sells to truckers"), get a first result in a few seconds and then send follow up instructions for each iteration (e.g. "add a testimonials section" or "make the background of the page a funky linear gradient")

We cared about time to first render: how soon after a command does visible output appear on the screen. It took some deep technical frontend engineering to stream the results from large language models to a running React app - so that users saw results instantly.

Overall, this was a notable speed up from traditional ways of building something: coming up with a design, writing the equivalent code yourself or dragging and dropping in a low code tool, and deploying your project. It impressed early users who knew how much work had now been abstracted away:


A real "in the wild" user review

But for large projects it still felt too...well...slow. You need to wait for the large language model (LLM) to make the next update, sometimes up to 30 seconds. Streaming results and rolling out GPT-4-Turbo helped, but it still didn't feel interactive enough.

With the new real time experience, you're no longer waiting for the LLM to complete before you can keep going. You just keep typing and occasionally check back in on the AI results. Alongside, the LLM examines the differences from your last app, last description, and current description, and bakes them into its next version of the app.

It feels much more equivalent to pair programming or collaborating with a teammate.

More aligned

Ironically enough – more speed per iteration has resulted in more time building overall. Early users submit hundreds of iterations:

1,000,000 lines of code generated

We find that you want to persist because it's also much easier to get exactly what you want.

As you build up your description of the app's UI and behavior, you're benefitting from the large language model's ability to guess how that description translates to code under the hood. Often that guess is shockingly accurate.

But what happens if it's not exactly what you had in mind? No problem. You just add more details – but only on the parts that need it.

High school math formula for the fans (Made with Create)

Since every version of the app is a function of the previous version and the latest changes to your description, you only need to add more detail to the things that aren't quite right. You're building up the minimal comprehensive spec to your result with code generation filling in the gaps that aren't specified.

Still, LLMs make mistakes. So we've also built in fast ways to jump back to previous iterations (hit cmd-z) and continue on from there. Fast recovery breeds confidence. You end up wanting to push the bounds knowing you can always go back.

The result is a much more immersive experience. It's easy to enter a state of flow, giving the large language model more and more information to get closer to your results.

So what can be made?

One of the fun parts about building Create is that we're constantly getting surprised at the bounds of what you can build. The best way to explore them is to build something yourself. English is the programming language so anyone can do it.

But if you want some inspiration, here are some early favorites (you can grab your own version of these with one tap):


Todo app (


A chat UI


Tic Tac Toe (

Rock, paper, scissors
A rock, paper, scissors game
Waiting list
A modern waitlist
Countdown clock
A live clock
Character converter
A character converter

Building a custom UTM tag tool for your business

So what will you build?

We're scratching the surface on who can build and what can be built in real time – everything from prototypes to games to demos to tools. Jump in and join the fun.

If you want help coming up with your own prompts, join our Discord community.

This is just the start.