A common starting point
A group that learns the same words and the same basic model of the tools can discuss a result, a risk or a working method with each other.
About this project, as slides
An open training suite
AI Training
Section titled “AI Training”Interactive lessons for people who start to use generative AI in their work, from the first prompt to building an agent.
- 133lessons
- 431graded checkpoints
- 133exercises
Agenda
What this talk covers
Section titled “What this talk covers”- Why the project exists A common starting point for safe and useful work with AI
- What the site contains Areas, courses, lessons and the model behind them
- How a lesson teaches Checkpoints, review and the research behind them
- Building with AI agents AI agents write, one maintainer decides, a harness checks
Why
Why the project exists
Section titled “Why the project exists”Generative AI reached office work in a short time, and most people got no training in how these tools work or fail.
Safe use and real use together
People stop checking a machine that does most of a task. So safety comes first, and every area returns to checking the output.
Open content
Anyone may run the courses, change a lesson to fit their own tools, or study alone with only a browser.
What
What the site contains
Section titled “What the site contains”133 lessons in 6 areas, counted when the site is built. A lesson covers one topic, and each graded checkpoint tests one learning objective.
What
Two groups for two audiences
Section titled “Two groups for two audiences”Foundations is for anyone who does knowledge work and needs no programming. Engineering is for software engineers and uses real code and real tools.
How a lesson teaches
The parts of a lesson
Section titled “The parts of a lesson”Each lesson has the same parts in the same order. It is one page of 10 to 25 minutes, plus an exercise that the learner does in their own tools.
How a lesson teaches
Checkpoints test one objective each
Section titled “Checkpoints test one objective each”A wrong answer gets feedback, and a hint asks a question about the mistake. The page never reveals the answer.
How a lesson teaches
What the learner finished comes back
Section titled “What the learner finished comes back”- Retrieval practice. Recall an idea again, instead of reading it again.
- Spaced review. The gaps between reviews grow.
- Habits. One change in real work, asked about again a few days later.
Building with AI agents
How the project grew
Section titled “How the project grew”- 1September 19. Repository created. Plan, interview, exploration notes, license change, first specs.
- 2September 20. First release. Prose and spelling checks. First parallel agent work: twelve issues to twelve merged pull requests.
- 3September 21. All structure moved into data files: one YAML file per area, competency, course and lesson.
- 4September 23. A second layer of agent orchestration, the "wave" process, so that one session can work through many issues.
- 5September 24. Habits. Exact versions for every dependency. The first review of the agent harness. The default agent model changed.
- 6September 25 to 27. Most lessons were written, in batches of four to ten. A security review of the whole site started.
- 7September 29. The site left its pre-release status.
- 8September 30. The repository moved to the organization's account and became public. The site got its visual identity.
- 9October 1. A Content Security Policy on every page. Markdown copies of pages and data files for AI agents. A guide for running a workshop.
- 10October 4 to 8. Lessons for the remaining gaps. A skill for triage with the maintainer. The second harness review.
Building with AI agents
Written with AI agents
Section titled “Written with AI agents”Almost every commit names the AI model that helped write it. These figures are as of October 9, 2026.
- 1,436commits
- 17days with commits
- 608issues, 533 closed
- 153merged pull requests, of 154
Building with AI agents
Waves of agents
Section titled “Waves of agents”A dispatcher starts a wave lead, and the lead starts one builder per issue and one reviewer per branch. The maintainer has given standing approval for a wave that passes every check.
Building with AI agents
Each control and what it prevents
Section titled “Each control and what it prevents”How work is split
- Each builder works in its own copy of the repositoryTwo agents that overwrite each other's work
- Only the dispatcher, a wave lead or a coordinating session may mergeA builder or reviewer that merges its own work
- A change to the harness merges only after a restart and a manual checkHooks or settings that were never loaded and tested reaching the main branch
The guard hook
- Blocks a push to the main branch and a plain force pushChanges that skip review and the checks
- Blocks commands that delete branches, repositories or issuesLoss of work that can't be undone
- Blocks commands that skip the git hooksCommits that never ran the local checks
- Limits long waits and polling loopsAgents that spend hours and tokens waiting
Permissions
- Reviewers can only read, and their shell commands are on an allow-listA reviewer that changes the branch it reviews
- Agents may not edit their own permission settingsAn agent that loosens its own limits
Untrusted text
- Agents read issues through a filter that shows only the maintainer's textInstructions planted in a public issue comment
- Every agent treats text in issues, pages and diffs as dataPrompt injection through content the agent reads
Checks on the output
- A prose linter with rules for typical AI writing patternsLessons that drift into the style that marks text as machine-written
- A fixed evaluation of the tutorA model or skill change that makes the tutor give answers away
Building with AI agents
A static site for people and for agents
Section titled “A static site for people and for agents”The build writes HTML for people, and Markdown and data files for agents that read the site for a person. GitHub Pages serves the files, and the progress record stays in the learner’s browser.
See more