The course: GS Core
The open path into the method. Sixteen short theory lessons intercut with five hands-on labs. You do not need the course to benefit from the work (a readiness assessment or remediation puts the method on your codebase for you), but if you want to practice it yourself, this is the way in.
What you finish with
A small project running under Generative Specification:
- A spec written before the code. The WHAT, stated as business rules and testable acceptance criteria, before the machine builds anything.
- A sentinel that routes. The root file that tells the assistant what the system is, which standards apply, the constraints, and where to read next.
- A gate that stops the build. A non-LLM check that turns a written rule into something the build can refuse.
- The rubric applied. Your spec scored against the seven properties, by a reader that has only the spec.
Who it is for
Developers and tech leads who already work with an AI assistant and want the discipline behind it. Bring a small project, or use Cancha Libre, a small public court-booking app you download and repeat step by step: github.com/jghiringhelli/cancha-libre. Each lab closes with one question to ask your AI, and you can run the same steps on your own codebase.
Bringing a legacy codebase? That has its own path in the workflow recipes; the course builds the muscle on a small project first.
What is published today
| Item | Status |
|---|---|
| Course index | Published |
| Lab 1: Turn the order around | Published (steps, prompts and the spec file from the recorded session) |
| Labs 2 to 5 | Written as scripts; dedicated lab pages ship alongside each lab video |
| Lesson videos | Being published. No lesson video is linked here until it is live |
| Field testimonials (video) | Published |
The lesson scripts are written in Spanish first; the English edition is in preparation.
The syllabus
Lessons are marked C (theory) and P (practice lab).
| # | Title | What it covers |
|---|---|---|
| P0 | Prepare your project (optional) | Pick your lane: follow along on Cancha Libre, or prepare your own project |
| C1 | The $327 million contract | The Mars Climate Orbiter loss as a case of a contract nobody could ratify |
| C2 | The inversion | Build top-down: write the what, let the machine derive the how |
| C3 | Why only now | Spec-driven development is old; what changed to make it viable |
| P1 | Turn the order around in ten minutes | Lab: the same request with and without a written spec |
| C4 | The discipline that combines | What GS takes from waterfall (rigor) and from agile (iteration) |
| C5 | The substrate | The retrieve, generate, verify loop and what the assistant stands on |
| C6 | The loop and generative execution | The AI brings up the live system and checks a use case layer by layer |
| C7 | Phase collapse | Specify, implement and verify stop being separate phases |
| P2 | Make the AI test your live app | Lab: generative execution on a running app |
| C8 | Leave nothing to chance | Why “the AI cuts corners” is often an under-specification problem |
| C9 | The ratchet | Every miss leaves something permanent; a defect is a query to the spec |
| P3 | One MUST, one gate, one turn of the ratchet | Lab: turn a written rule into a check that stops the build |
| C10 | The bridge and the asymmetry | Why the model follows a structured spec, and why reading is easier than writing |
| C11 | The disciplines that activate | Name SOLID, hexagonal, TDD in the spec to invoke what the model already knows |
| C12 | The sentinel | The root file: what belongs in it and why size matters |
| P4 | Your sentinel in fifteen minutes | Lab: write and test your root file |
| C13 | The layers that rise | One spec, verified differently at each environment |
| C14 | The rubric | The seven properties and the stranger test |
| P5 | Score your spec | Lab: a reader with only your spec scores it |
| C15 | What you’re buying | What changes in value and in cost when the specification is the work |
| C16 | What is left to you | Specify, generate and verify are one movement; what stays human is turning a conversation into a complete spec, and signing off on the evidence |
After the course
- The rubric: the seven properties in full, with the failure named for each.
- Spec completeness: when a spec is complete enough for a stateless reader.
- Quality gates: the non-LLM checks that make it enforceable.
- Workflow recipes: the method on a real project, greenfield or existing.