Method

How we build

Phase gates, a shared platform, and a kill decision written down before anyone writes code. Speed comes from discipline, not from heroics.

The short version

Every company starts as a spec. The spec names the user, the job to be done, the platform packages it will reuse, and the conditions that end the idea.

Phase 0 is that spec. If the platform does not fit, or the problem is not a company, the work stops there. Stopping early is a feature.

Build is a small senior team. Specs stay in front of the code. AI drafts, tests, and reviews alongside engineers. Engineers still own every line that ships.

Spinout is the point. The company takes its own name and its own data. The studio keeps the platform and gets faster.

Phase gates

Four stations. Two exits.

Every station ends with a written question. If the answer is no, the idea stops there.

  1. Station 01

    Idea

    A painful, specific problem that deserves a company. We keep a long list and say no to most of it.

  2. Station 02

    Phase 0

    One spec: the user, the job, what the platform already covers, and the written conditions that kill the idea. Most ideas end here, cheaply.

  3. Station 03

    Build

    A small senior team ships on the shared platform. Real users early. Weekly demos, not quarterly decks.

  4. Station 04

    Spinout

    The company takes its own name, its own team, and its own data. The studio keeps building the platform for the next one.

GL-PROCESS · scroll-driven

Gate 01
Is this a real problem for a reachable customer?
Most ideas stop here, before any code.
Gate 02
Does the spec hold, does the platform fit, and are the kill criteria clear?
If not, we stop and keep the notes.
Gate 03
Are real users choosing it, and is it getting cheaper to serve?
Gate reviews against the written criteria.
Gate 04
Can it stand on its own name, team, and data?
Then it leaves the garage.

Operating principles

Six rules we don't bend.

  • 01

    Write it down first

    A spec beats a meeting. If it is not written, it is not decided.

  • 02

    Name the kill switch

    Every idea carries the conditions that end it. We read them out loud at every gate.

  • 03

    Reuse before you build

    If the platform already does it, the company does not rebuild it. Ever.

  • 04

    Ship to real people

    Demos are for customers, not for us. The first user shows up in weeks, not quarters.

  • 05

    Small teams, senior hands

    Fewer people, more ownership, AI in the loop. Headcount is not a milestone.

  • 06

    Spend like it is ours

    Lean infrastructure, lean CI, no vanity tooling. Every dollar saved is runway for the company.

The platform

Built once. Inherited by every company.

Core

Shared platform

  • Identity
  • Billing
  • Records
  • Admin
  • AI orchestration
  • Forms & approvals
  • Company 01
  • Company 02
  • Company 03
  • Company N

The core is built once and versioned like a product. Each new company starts ahead of a blank repo, and every fix flows back to the next one.

GL-PLATFORM · live diagram

The stack

Boring where it should be. Sharp where it counts.

  • Python

    Where the models live: data, evaluation, and AI workflows.

  • Rust

    Where it cannot fail: the orchestrator and the hot paths.

  • Shared packages

    Versioned, tagged, and pulled by each company. Never copied.

  • Lean CI

    Fast checks that run when they matter. Costs watched like a feature.

“A studio earns trust by naming what it will stop.”

GarageLabs operating principle

The future is being built right now. Build it with us.