The first release is a decision
Published 27 September 2026. This is the web version of the email.
This is the first Graylemon Journal by email. We're an AI-native product and venture studio, and we run our own ventures through the same method we bring to every build. Each issue collects what we've published since the last one, with the part worth your time pulled out, so you can skip the rest.
This one is about the first release: what goes in it, what stays out, and why the second list matters more.
The guide: how to scope an AI product before you build it
Scoping is seven decisions made in writing before anyone designs a system: who it's for and which moment it changes, what success looks like and what would make you stop, whether the data exists, which parts need a model at all, what the first release leaves out, which constraints can't move, and how you'll know it works before everyone sees it.
The one that gets skipped is the stop line. Write down the result that would make you stop while you're still free to believe it. After launch, every number starts to look like a reason to carry on.
The Lab: Cut It
A dental clinic, a budget of five build points, and six features that all want them. Sort each one into build, later or cut, then see our call and the reason for each. Nothing you choose leaves your browser.
Play it before you read our answer. The useful part is noticing which feature you argue for.
The framework: Decide · Design · Ship
Three steps, one team, no handovers. Decide what's worth building and what gets cut, design the screens people will actually use, and ship it tested, staged and observable.
The misread to watch for is designing before deciding. Screens make an undecided scope look finished, and a finished-looking feature is much harder to cut.
The series: Stuck-to-Scale, stage by stage
Six build logs, one for each stage a venture moves through. Each asks the one question its stage has to answer before the next can start.
- Ideate: is the problem real?
- Define: what exactly are we building?
- Build: does it exist, and does it work?
- PMF: do they keep using it?
- Market: can we reach them repeatably?
- Grow: does it compound without us?
One call
In apprn, our agentic salon management venture, the product's spine is one verified event, not a model. An artist can only start a service by entering the one-time code sent to the customer with that booking, and pay follows that record. AI goes only where judgement helps.
Whether it holds is decided in a live test with real salons, against kill criteria set before the test began. We'll publish the criteria before the result.
A question for you
What's the feature you'd find hardest to cut from your own first release? Reply and tell us. We read every reply.
Graylemon