My (small) frustration with "Bring!"
Yesterday morning I got annoyed with the actually quite nice app "Bring!". In case you don't know "Bring!": it's a shopping list app. You can have a number of shopping lists, and add products to each list, together with others, and while shopping you can check off the products with one click. The products are represented by small "tiles." Very practical. Unfortunately, though, it is increasingly cluttered with ads - flyer tabs, ingredient advertising, and when you start typing a product, products from certain brands are often the first hits.
Besides that, I find the app a little stone-age by today's standards. There is an integration with Chefkoch, and that's about it.
So: in times of vibe coding, I had to create my own shopping list app. Unlike Bring!, I wanted to preserve the shopping list metaphor, and I wanted strong AI integration: you photograph the ingredients of a recipe, and an LLM puts them on the shopping list.
No sooner said than done:
8:00: Wrote down the specification. 160 lines long. The document only describes the concept and the desired features.
8:30: Discussed the specification with Codex + GPT 5.6 Sol (high). Cleared up a few ambiguities, ChatGPT updates the specification
8:45: used /goal to let the LLM get started.
11:00: after pressing "okay" a few times: app is finished.
11:30: had two styling blemishes fixed via "here is a picture, make it prettier," pointed out a small bug (shopping does not reset the quantity, i.e. you add 5 tomatoes, check them off, add 3 tomatoes, and end up with 8 tomatoes) and had it corrected.
11:50 The app is running on my VPS. I'm not good at that, but I handle the ssh on my VPS by hand, even if I don't understand the nginx stuff.
My favorite feature is the AI recipe photographing: click "Photo," take a photo of the ingredients in a book, on a website, wherever. LLM (more precisely GPT Terra) analyzes it, the app suggests ingredients, you can deselect or change them, done.
And the craziest bit: the recipe image is exactly what the LLM got. And it reconstructed the "Kapern" in the title correctly.
The next day I still had to - I never would have thought of it - change the nginx setting so that it doesn't only allow tiny 1 MB images through.
Then quickly had one more feature developed so that the app remembers the shopping frequency of all products and, for a weekly grocery trip, suggests what I should be thinking of.
Result
My app is exactly what I wanted, a shopping list the way I need it, shared with others, can understand any recipe, knows the things I always have at home, the things I buy regularly. With lots of nice details.
It cost me 40% of my ChatGPT Plus weekly quota, so not too much.
Runs on phone, tablet, computer.
Quite impressive
Where previously only very small apps were within reach of the "I'll build that myself" possibilities, now even more extensive apps like "Bring!" are actually unnecessary. I'm not saying that everyone can have such apps built - writing a clean specification is not easy, and thinking about security aspects isn't either. Nevertheless: especially now that the LLM providers are also offering ways to host software built with their own LLM, I feel confirmed in my opinion that apps, especially paid apps, are already almost a thing of the past today, and in 1-2 years irrelevant.
Too bad for app developers... And you wonder what comes next?