Level three of a Gen 2 studio: build what you need

I have run studios for twenty-five years. I am not a software engineer. This year I started building the software my studio needs, and I want to be precise about why that is possible now and what it does and does not mean.

The tools every studio wishes existed

Every studio has three or four things it does its own way, and off-the-shelf software either does not have them or has a version that does not fit. Mine were an instructor bonus model that rewards the classes that actually fill, measured against the true capacity of the time slot rather than the size of the room. PT commissions. A small-class notice so a teacher knows before they walk in. An intro-pass tracker that flags who is about to lapse before the pass runs out. A Meta tracker. An invoice creator.

In a Gen 1 studio the path to any of those was: problem, request, specification, roadmap, developer, integration. Six steps, most of them someone else's, and a year later you have a feature that almost does what you asked. The Gen 2 path is: problem, describe, build, use.

AI writes the code. Experience decides what should exist.

The code is not the hard part any more. Claude Code, Codex, Gemini and the rest will write it. The hard part is knowing what should exist, and twenty-five years of standing in studios is the only thing that answers that. A software engineer would not know that a 6am slot that realistically holds twelve should be judged against twelve, not against a room built for twenty-five. I do, because I have watched compensation punish good teachers for the timetable for two decades.

Build on the business, not another silo

Here is the part that makes it Gen 2 rather than a hobby. Every one of those tools is built on the business context that already exists underneath: members, classes, teachers, payments, operations, read from the systems the studio already runs. Nobody exports a CSV. Nobody re-enters anything. The instructor bonus model reads the same attendance the owner has been asking questions of at level one and the same schedule the instructors see in Tribe at level two.

That is what "build what you need" means. Not that AI can build apps, which everyone knows by now, but that the people closest to the problem can build on a shared understanding of the business instead of beside it. Software at the speed of the operator.

If you want to see exactly what one of these looks like, the Instructor Bonus Builder is published on GitHub with full documentation, including a fully worked month, and a live demo that needs no login. You can run it on your own studio or gym, or build your own from it. A deterministic engine splits a fixed pool on retention and capacity, reconciled to the cent, and every allocation opens a panel that explains the number line by line. The AI explains the maths. It never decides who gets paid what.

What this level is, honestly

These are examples built on the business context, not included modules, and not a catalogue. What you can build depends on your systems, your data and the job you want done. So level three starts with a conversation: tell us the job, and we will tell you what is possible now, what needs configuration and what still needs work.

Where it sits on the path: you get here when a tool you wish existed comes up, not before. Build one, use it, then decide on the next. And once the business is understood well enough to build on, it is understood well enough to hand work to, which is level four.