How AI Works
The AiGovOps Foundation's planned free course on generative AI: one ladder from complete beginner (100) to research depth (500), with every lesson in three gears and a public, checkable truth standard.
Status: Level 100 published (18 Nov 2026); Levels 200 to 500 planned. Level 100's 11 modules passed the truth standard and are on the course site. This README and the other planning documents are still drafts. Plan: How AI Works 100–500: the agent-run build plan.
The live Foundation Academy offers Learn 101–104 and Practice 100–400. This course's 100–500 ladder is a build plan, not a claim that the Academy has already replaced those offerings. The shared thread is Get to Yes / Stay at Yes / Recover to Yes. Foundation membership is optional; the course is free to read. These are not university degrees.
The shape
- One ladder, five levels. 100 How AI works · 200 Using gen AI well · 300 Building with gen AI · 400 Engineering AI systems · 500 Governance, safety and research. See CURRICULUM.md. Credentials: Open Certificates (100–300), Open Mastery (400) and Open Advanced Mastery (500); see credential names and design.
- A placement test routes each learner by what they can do, not by credentials (specs/PLACEMENT.md).
- Three gears per lesson. Plain (story, no math), standard (worked examples), technical (equations and papers). Same objectives and the same quiz in every gear.
- Five strands at every level. How gen AI works · building with it · governing it · verifying claims · a project. The governing strand is the Foundation's practice ladder.
- A truth standard you can audit. No citation, no publish. Two human reviewers per module, a red-team pass, a board sign-off per level, and an errata log on every module. See TRUTH-STANDARD.md.
- Agents do the work; people sign. Thirteen agent roles draft, check, build and run the course. No agent ever counts as a reviewer. See AGENTS.md.
Where things are
| Path | What it holds |
|---|---|
levels/<n>/objectives.yml | Every module's objectives for that level: the single source the quizzes map to |
levels/<n>/modules/<id>-<slug>/ | One module: module.yml, plain.md, standard.md, technical.md, quiz.json, workbook.md, sources.yml, review.yml, ERRATA.md |
levels/<n>/board.yml | The review board's sign-off for the level |
templates/module/ | Copy this folder to start a module |
tools/check_modules.py | The truth checks CI runs on every push |
agents/ | The brief each agent role works from |
specs/ | What the builder agents build: placement test and lesson viewer |
STATUS.md | The program lead's scorecard |
Run the checks
pip install pyyaml
python3 tools/check_modules.py
python3 -m unittest discover -s tests
Build the draft preview
Run python3 tools/build_site.py, then python3 -m http.server --directory _site. The site includes only selected planning documents, the objectives map and the real course apps; tests, fixtures, templates and tooling are not published. Empty apps are not linked from the landing page. The Pages workflow deploys after checks on main (or manual dispatch from main). Bob merges; Devin Review and Claude review remain the Builder gate. Enable GitHub Pages with GitHub Actions as the source in repository settings before the first deployment. A feature-branch push does not publish the preview or change repository visibility.
The preview uses the Foundation's Wren Café colours and loads its Fraunces, Source Sans 3 and IBM Plex Mono fonts from https://www.aigovops-foundation.com/fonts/fonts.css. This contacts the Foundation for font assets, not analytics. System fonts remain the fallback when offline or when that stylesheet is unavailable. Standalone viewer/placement builds remain self-contained.
Licences
Foundation-owned course content in new releases: CC BY-NC 4.0. See the licence for written permission for paid use, earlier releases, and third-party terms. Code in tools/, tests/ and the apps: MIT.