Series contents

Series

Building RepGarden

Build series · Discovery

Discovery.

Two calls with Benji, a list of questions and nothing built yet. Why the build starts here and not with a prompt.

By Ahmed Gadir3 min read

You could have it by tonight

Benji is a personal trainer and strength and conditioning coach. He wants an app to use with his clients: their programmes, their workouts and their progress in one place, instead of the tool he pays for now and finds clunky. I could describe that to a coding agent this afternoon and have something that looks like it by tonight.

That is the problem with it. The first version of almost any app is easy to get now. A beautiful one, in minutes, from a paragraph. The question is what you would actually have.

The four parts of a great app

A great app does four jobs at once. It solves a real-world problem for particular people. It solves the technical problems of making that work and keep working. It creates something that looks and feels right. And it communicates why anyone should care.

The four parts of a great appA wheel of four quadrants: 01 solving real-world problems, 02 solving technical problems, 03 creating art, 04 communicating value. A small accent sector inside each quadrant shows that AI covers part of each on its own, not the whole.0102030401Solving real-world problemsFor particular people, with a problem worth solving.02Solving technical problemsMaking it work, and keep working.03Creating artThe look, the feel and the character.04Communicating valueWhy anyone should care, said clearly.
The four parts of a great app. The filled centre of each quadrant is roughly how far AI gets on its own: some of the way on each, all of the way on none.

AI can help with all four, to a degree. It will write the code and it will draw you a logo. Put it fully in the driver's seat, though, and you get generic work, because it has never met Benji. It does not know him or his clients. It does not know what they actually want from a trainer, or what makes them stick with one. It can give you a thousand options. It cannot tell you which one is right, because the answer is not in the model. It is in the people the app is for.

Step one: discovery

So the build starts with Benji, not with a prompt. Discovery means finding out three things before deciding anything about the app:

  • What he uses right now, and why it is not working for him.
  • What his ideal app would do for him and for his clients.
  • A few things about his business that most people would skip.

The third is where the useful answers are. The questions I took into the first call were about his clients rather than about features: who they are, why they come to him in the first place, what they are really after when they sign up, and what stops the ones who nearly sign up and don't. Any founder with an idea faces the same four questions, and most skip straight past them to the feature list.

We got through the first two on the first call. The other two need a second call, which is booked. In between, I go away, do some research on what his clients and other trainers already use, and come back with a proposal.

Where the build isFive stops in a row: Discovery, Research, Proposal, Build, Live. The build is at Discovery.DiscoveryResearchProposalBuildLive
Where the build is. Discovery first, then research, then a proposal Benji can say yes or no to. Nothing is built until that point.

What Benji has left open

Benji has given a lot of creative freedom on two things: the branding, and the gamification, the part of the app that makes coming back feel like progress rather than homework. Both are decided later, once the research is in, and both will have their own entries. For now they are open, and that is deliberate: deciding the look before knowing the clients is how an app ends up pretty and unused.

Where things stand

Nothing has been built, and nothing should have been. One discovery call is done and a second is booked. The next step is research, then a written proposal. The project's code and notes are public in the RepGarden repository on GitHub.