Imagine if we treated courses and educational material the same way we did with code.. completely reusable/ composable/ shareable. Growing reflections of our interests and ideas.
We are in a new era of coding where agents can completely control the end to end software development pipeline from ideation to spec/ planning to testing and iterating.
What if we utilized their same capabilities for course creation and education? Yes i know there are so many edtech wrappers on llm api’s but imagine if they were truly harnessed for education..
And those same courses could be used/ updated/ referenced/ customized by other humans, ai agents and tools. We can easily do this with MCP and CLI as tool surfaces — exposing the same underlying tools for course creation for both agent and human.
Imagine if you could interact with all of the past courses you’ve taken — those distant courses you took a random summer or that Predictive Statistics class in Grad school. All through the same interface… And you could update them! With access to an MCP, you could routinely task your agent (Cowork, codex, Hermes, OpenClaw) to scrape the internet for newly released academic papers and update the course accordingly.
Imagine a dynamic course that grows with you!! You can interact with your course material based on different jobs you might have: “Explain how I might utilize statistics in my new job as a data analyst”. Your agent already has context about your role, but also can query what you learned in your stats class and reframe it to your current position/ situation.
All of these walled-garden textbook providers and course catalogs are being threatened by tools that were trained on them… Imagine a world where anyone can learn anything they want, howEVER they want.. and they can take it with them. and update it. it evolves with them, with the times… You watch a super in-depth video on Youtube and suddenly you can spawn 8-10 subagents in parallel that search for sources, plan out a course, and curate a course based on citations! This is the future of learning.
But how?
Okay — every course can be treated like code - simple markdown format with frontmatter YAML at the top to provide some descriptors like course name, duration, prerequisite, date, and sources. Version controlled with git, maintained like any other codebase. Every agent already knows markdown. There’s a reason certain agentic memory solutions have emerged that all use markdown files and a filesystem — OpenClaw, LLM wiki, OKF, etc.
Then, that course (which is really just code) is ingested to a backend via CLI or MCP that the agent has access to. Chunked, embedded, and stored for retrieval.
Then, those courses are rendered in the browser and the SAME data is exposed to both the human and the agent. The human sees the pretty rendered markdown in browser with all of the mermaid diagrams, syntax highlighting, runnable code, and markdown while the agent can retrieve the super friendly markdown and chunks that they know so well. Once those courses are chunked and embedded, an agent can do whatever they want with it:
- create study guides
- generate flashcards
- create quizzes
- generate audio overviews
- draw connections between various courses
As the agents gain new tools and capabilities, so too does your learning.
Finally — the nice thing with all of the same course creation/ interaction tools being exposed to the human and the agent allows you to interact with your courses using whatever tools you already use (Claude Code, Cowork, Codex, Cursor, Gemini CLI, Hermes, OpenClaw).
Humans need to evolve towards the next era of learning, using the same tools that is skyrocketing agentic coding progress.The internet is the source, we need the proper tools to learn from it and harness it.
That is what I am building https://www.learnosapp.com to try and solve…