← All field notes TALK · No. 002

I Bet AI Couldn't Replace Developers, So I Made It Build an App

I bet a friend that AI can't replace developers, then put it to the ultimate test: building a fully automated grocery-saving app without touching a single line of code.

Mateusz Podeszwa · June 15, 2026 · 2 min read

The Zero-Code Bet: Testing the Limits of AI

I recently made a bet with a friend. He said AI is going to replace all programmers. I disagreed.

I do not think current AI can replace a developer's ability to reason through problems, spot patterns, and make good architectural decisions. To test that properly, I decided to build a new app under one strict rule.

I would use a programming language I do not know, and I would not touch the codebase myself at all.

The Idea: A Real Grocery Saver

I have had this idea in mind for a while. I wanted an app that actually helps people save money on groceries they buy regularly. Not something like a basic discount app that tells you a random item is on sale, but something genuinely useful.

The same or very similar products can vary a lot in price between shops. It could be fruit, rice, pasta, or almost anything else. If you want to buy strawberries, one shop might sell them for £4, another for £2, and a corner shop might charge even more.

Most people do not know those price differences, especially if they do not have time to compare stores. I wanted to explore an app that could tell you where to buy the cheapest ingredients around you without making you do all that work yourself.

How It Works: Enter "Sage"

The workflow I designed for the AI-built app is based on OCR and AI processing.

  1. Scan: You take a photo of your grocery receipt.
  2. Process: The system reads the receipt, identifies the products and prices, and stores them in a database.
  3. Plan: Later, you ask the assistant, which I call Sage, something like, "I want to make tomato soup."
  4. Save: Sage checks the stored pricing data, finds the cheapest shops for the ingredients, builds a shopping list, and estimates the total cost.

The Flaw, and Why I Still Built It

The idea has a very obvious weakness. It depends on people actually scanning receipts.

That is a serious assumption. Some smaller shops do not always provide usable receipts, and even when they do, many people will not bother uploading them. That creates a data collection problem that could easily limit how useful the app becomes.

Even so, I still wanted to try it. I put no manual coding effort into the build itself, so the risk is low. If the project fails, it proves something about the current limits of AI-assisted development. If it works, then it could turn into something genuinely useful. Either way, the experiment is worth it.