Five Hundred Guides, From First Principles to Agent Harnesses
By glitchdata Team 2 min read
The guide library now holds 500 articles. They were written to fill the gap between a course lesson, which has to stay on one path, and the documentation for a tool, which assumes you already know why you are reading it.
Sixteen Topics, Two Ends of the Field
At one end sit the subjects that have not changed in a decade: how machine learning actually works, the statistics behind an evaluation, how to shape data before anybody trains anything, and the engineering that moves it around.
At the other end are the ones that did not exist in their current form two years ago — retrieval-augmented generation, prompting as a discipline, the harnesses that let a model use tools, the Model Context Protocol, agent-to-agent communication, what people mean by AGI, and the security problems that arrive with all of it.
Each topic is a filter on the index, and each guide carries a cover drawn from its topic rather than a stock photograph, so a list of fifty articles still reads as fifty distinct things.
Written to Be Read Once, Properly
A guide is not a summary with links to somewhere else. The body runs the full width of the screen with its contents beside it, so a long piece stays navigable, and the whole article is in the page — there is no account to create and no second half to unlock.
They Point at the Courses, and the Courses Point Back
Guides are not a separate library bolted on beside the courses. A lesson can recommend specific guides as further reading, chosen by the person who wrote the lesson; when nobody has chosen any, the lesson falls back to guides on the course's own topic. Each guide page lists the lessons that recommend it, so a guide found through search leads back into the structured path it belongs to.
If you would rather read them somewhere else, the guides feed carries the newest thirty, and the API will return any of them as JSON.