The Last 20% Is the Whole Company
AI made the first 80% of any app cheap. The expensive part never moved: identity, billing, telemetry, provider failover, support, rollback. Building one product per theme meant paying it five times — so we built it once and named it Rumble Rails.
Something flipped in the last two years and most companies have not reorganized around it yet.
The first 80% of a software product — the screens, the data model, the happy path — got cheap. A domain expert with an assistant can get a working prototype in a weekend. We have done it repeatedly.
The last 20% did not get cheaper at all.
What the last 20% actually is
Make a list of everything a product needs before a stranger can pay you for it and keep paying you without emailing you:
- Identity that works across web, native, and extensions, with roles and row-level security so one tenant never sees another.
- Billing that handles seats, plan gating, trials, failed cards, and usage meters — and does not drift from what the UI says is allowed.
- Telemetry that shows you a problem before a support ticket does.
- Third-party providers — LLMs, SMS, email — with budgets, timeouts, and a failover path, because every one of them will have a bad afternoon.
- Support with case history, device diagnostics, and a way to find out what the user actually did.
- Rollback. For code, for data migrations, and especially for anything with a model in it.
None of that is a feature. None of it demos well. All of it is the difference between a prototype and a product someone trusts with money.
We rebuilt it five times before we built it once
I will be honest about the order of operations. We did not start with a platform. We started with products — CapturedIt, then Premodoro, then HuniDu, Bread & Butter, and nMyCity. Five themes of a person's life — memory, focus, home, work, community — that look nothing alike as software.
Somewhere around the fourth billing integration we stopped and looked at the shape of what we had written. Same auth pattern. Same entitlement checks. Same Stripe webhooks. Same "where did the error go" dashboards. Same provider timeout bug, fixed four slightly different ways.
So we pulled it out. That layer is Rumble Rails — the production stack under every Rumble theme, and the one external teams can deploy on through Rumble Built Services. It is how we build; it is not the product.
Why a map instead of a roadmap
Platform claims are easy to make and hard to check. So on the platform page we drew it as a transit map: lines are capabilities, stations are the public products that actually run on them, dashed track is under construction with a quarter attached.
The honest version is less impressive than the slide version. Premodoro runs fully on Rails today. The other four are being connected one at a time. The map says so. It will keep saying so until it is not true.
What AI has to do with it
Nothing — and that is the point. Every capability on Rails works with no model anywhere in the stack. Where a model is useful, it rides on the provider line like any other third-party dependency: with a budget, a timeout, a fallback, and a rollback plan.
AI is a component we govern. It is not the product, and it is not the platform.
If you are building right now
You will ship your first 80% fast. Budget your attention for the other 20% before you need it, because that is where the company actually lives. Or run on rails that already exist — ours or anyone's — and spend your scarce engineering on the part of the problem only you understand.