A $345 million app that you can replace with a single prompt

Hey!

This has been swirling around in my head lately:

MyFitnessPal generated approximately $310M in revenue in 2023, and in 2024 those revenues are estimated at around $345M.

In 2024, MyFitnessPal still had an estimated 85 million monthly active users who collectively log around 16 million different foods daily (industry estimate, not an official separate annual figure).

Here’s how I’m thinking about it:

The core function, taking a photo of food and getting calorie estimates, can now be done with a single ChatGPT prompt.

I’m not saying the app is worthless. The community, integrations, polished user experience, all of that has value. But the personalized result? The actual main output you’re paying for?

That’s just: (1) Input + (2) Rules -> (3) Output.

And that’s exactly what a good prompt does.

The Pattern Behind Every Simple App

Now, if we think about it, most useful apps follow the same formula:

Snap a photo of food, get calories
Paste messy text, get structure
Enter expenses, get a budget breakdown
…

The global nutrition apps market reached around $5 billion in 2023 and is now, in 2025, deep into a rapid growth phase. By 2033, it’s projected to reach approximately $14 billion. That’s a lot of money for apps that fundamentally do one thing: process your input through certain rules and return an output.

ChatGPT can do all of this.

The difference?

You write the rules once, save them, and they work forever.

Where We’re Heading

As you can see, every interface you use to access these most powerful AI models is gaining new “powers.” It’s now taken for granted that it can browse the internet, you can upload images, it can analyze them, recognize audio, draw charts… Google Gemini 3 excellently analyzes video too -> Try it 😉

So we can conclude that capabilities will only keep growing…

And you can use all these capabilities in a prompt that defines your desired workflow. I have to admit, I’m building a library of these myself. And I’ll be sharing it with you soon.

Yes, you could buy a subscription to a dedicated app with a well-designed UX flow. But you know what, that flow designed by the best UX expert sometimes doesn’t beat a simple conversation that I direct to my current situation. Predefined flows in apps are designed to capture the largest possible audience and aren’t adaptable.

So this can be a completely realistic perspective:

Instead of paying $12 here, $9 there, $15 somewhere else, you can move simple problems (simple mini apps) into ChatGPT, onto one AI subscription.

How to Build Your Own “App” in 10 Minutes

Here’s the key fact you need to consider in this case:

A prompt is not a question. A prompt is a program specification.

You’re not asking the AI a question. You’re telling it exactly what to do, step by step.

Example 1: Calorie Estimator

You are a “meal estimator.” When I send you a photo of food:

Identify visible ingredients.
Estimate portion sizes (grams, pieces).
Estimate calories, protein, carbohydrates, and fat.
Add this meal as a new row in my food diary table.
Update the weekly summary and display a weekly chart (calories and macronutrients).

For each response:

Show the updated food diary table.
Show the updated weekly chart.
Add an overall confidence level (low / medium / high).
Note what I could clarify to improve accuracy.

Rules:

Use only estimates and clearly state this.
If the image is unclear, say what’s missing.
This is not medical or nutritional advice.

Same output as the best app. Sometimes better, because you can ask follow-up questions.

Example 2: Text Structurer

Instead of a $12/month note organizer:

AGENT BEHAVIOR: When the user types START and presses Enter:

switch to content collection mode
do not process, summarize, or respond to individual inputs
silently add each subsequent input (even if multiple lines, interrupted by Enter, or spoken) to an internal record

During collection mode:

do not react to Enter
provide no output or feedback
do not modify, correct, or interpret the text

When the user types STOP and presses Enter:

stop collecting
process exclusively the content collected between START and STOP
use the following output structure:

OUTPUT:

Summary (2 sentences max)
Key decisions
Action items (who, what, by when)
Open questions

RULES:

do not mention START or STOP in the output
do not add explanations or additional context
if data is missing, mark it as “not specified”
output should be a clean, scannable list
be concise and consistent in format

Why This Way?

Three things happen when you build your own prompt-apps:

Cheaper. One AI subscription instead of 10 subscriptions.
Consistent output. Once the prompt works, it works every time.
Instant customization. Want a new field in the response? Change the prompt. No feature requests for app upgrades.

Try This Today

Pick one app you use every day. Write down what input you give it and what output you want. Turn it into a prompt using the templates above.

Test it.

You just built your first “AI app.”

What’s Coming

Simple apps will become prompts. Complex systems (banking, healthcare, complex tools…) will remain systems. But their “front door,” how we users interact with them, will increasingly move toward conversational mode.

Well, I hope I’ve managed to show you the possibilities that are opening up. And I’ve only touched on two examples here.

I’ve decided that in the next posts I’ll definitely share more on this topic with you.

Talk soon,

Primož

P.S. Reply with one app you’d like to replace with ChatGPT. I’ll send you a prompt template for it. 🤝