CMU's free CS Learn platform puts 4 full courses on YouTube, including an AI agents class. Here's what each covers and the order to actually watch them in.
Vamsi Tallapudi
Manager, Architect Technology at Cognizant
Carnegie Mellon just put four full semesters of its computer science classes on YouTube for free, through a new platform called CS Learn, and one of them is a graduate-level AI agents course taught by the co-creator of OpenHands. No application, no tuition, no login wall beyond YouTube itself.
The announcement landed on November 16, 2026, and the School of Computer Science has been rolling out one new course per day since launch week. I went through all four, checked the prerequisites against CMU's own course listings, and worked out the order that actually makes sense if you're starting from zero. Here's what's live, what's still coming, and which course to open first.
CS Learn is a free platform from CMU's Kenneth C. Griffin School of Computer Science that publishes real, current-semester lecture recordings to YouTube and to cs.cmu.edu/learn. These aren't dusty archived lectures from a decade ago. CMU is recording this semester's classes and releasing them within weeks, using a student group called ScottyLabs to run an AI-assisted production pipeline that edits the footage, blurs student faces, and strips identifying audio before anything goes public.
One detail I actually like: on-screen text captures the real questions students asked in the room, plus the professor's answer. You're not just watching a lecture get delivered at a wall, you're watching the actual back-and-forth.
Phillip Compeau, a teaching professor in CMU's Computational Biology department and one of the people behind the launch, said the platform lets CMU "share the teaching expertise of people on the frontier of their subjects." Four courses shipped at launch. CMU says more arrive starting in January.
Here's the full lineup, with the course code CMU uses internally so you can look up the syllabus yourself if you want more detail than a YouTube description gives you.
| Course | CMU code | What it covers | Who it's for |
|---|---|---|---|
| Principles of Computing | 15-110 | Python basics, algorithms vs. programs, data types, conditionals and loops | Anyone, zero CS background needed |
| Software Architectures | 17-335 | Designing and evaluating systems at an architectural level, not line-by-line code | Anyone who's written small programs before |
| Generative AI | 10-423 | Diffusion models, language models, generating images, text, music and art | People who already know ML or deep learning basics |
| AI Agents | 11-768 | Agent scaffolds, tool use, evals, coding and computer-use agents, training agentic LLMs with RL | People comfortable with LLMs and ML fundamentals |
Principles of Computing is the only one of the four with no prerequisite at all. Software Architectures assumes you can write code in some language but doesn't care which one, and explicitly doesn't require any prior data structures or algorithms coursework. Generative AI is where it gets real: CMU's own prerequisite list for 10-423 requires an intro machine learning course (10-301, 10-315, 10-601 or similar) or an intro deep learning course before you even start. AI Agents sits on top of that same foundation, then adds reinforcement learning on top.
Based on the prerequisites CMU itself publishes, here's the order that avoids you getting stuck three lectures into a course that assumes math you don't have:
Here's the catch nobody's mentioning: none of these four courses actually teaches you machine learning fundamentals. Generative AI and AI Agents both lean hard on ML and deep learning knowledge that CMU expects you to bring from somewhere else. If you're coming in cold, you'll want to fill that gap with something like Sebastian Raschka's from-scratch reasoning models book or an intro ML MOOC before you hit course three on this list. CS Learn isn't a self-contained degree path yet, it's four strong courses with a gap in the middle.
Out of the four, the AI Agents course (11-768) is getting the most attention, and for good reason. It's taught by Graham Neubig and Daniel Fried, and Neubig is one of the people behind OpenHands, the open source coding agent that's become a reference implementation for a lot of the agent tooling people build on top of LLMs today.
Back when the course was first announced for CMU's Fall 2026 semester, Neubig described the goal as teaching students to "create a scaffold, build evals, and train an agentic LLM using RL," while balancing theory with the frameworks people actually use in production. That's a genuinely rare combination for a university course. Most CS programs are still catching up to agents as a topic at all, let alone running a full semester with assignments built around RL training runs.
On CS Learn, the AI Agents course premieres November 19, a few days after the platform's November 16 launch. A Hacker News thread discussing the launch singled it out too, with one commenter saying the lecture content "makes you want to learn everything" about the subject.
That same thread had a split reaction worth mentioning. Several people asked whether learning CS fundamentals even matters anymore now that LLMs can write code for you. One reply pushed back hard: if you learn the fundamentals properly, AI becomes a power tool you control, not a crutch you're dependent on. I land on that side of the argument. Watching Neubig and Fried's course without knowing how a transformer or a training loop actually works means you're memorizing commands, not building anything you can debug when it breaks.
No, and CMU isn't pretending otherwise. You get the lectures, the Q&A captions, and whatever slides show up on screen. You don't get graded assignments, office hours, a transcript, or a CMU credential. If you want the full experience, including homework and projects, the AI Agents course's own site at cmu-agents.com has a public syllabus and assignment repo you can work through on your own, separate from the CS Learn video feed.
What you do get for free is something that didn't really exist at this scale before: a current-semester, name-brand CS curriculum on a platform you already have open in another tab. Compare that to learning platforms built around AI tutoring, which solve a different problem, pacing and feedback, rather than giving you the actual lecture content from a specific university course.
Subscribe to CS Learn at Carnegie Mellon on YouTube, or go straight to cs.cmu.edu/learn and pick a course from there. Each course is organized as a full playlist, one video per lecture, in the order the class was actually taught. There's no sign-up form, no email capture, nothing. You click play.
If you're only going to watch one lecture before deciding whether this is worth your time, start with the first Principles of Computing lecture below. It's the clearest example of what CS Learn actually looks like in practice, blurred faces, captioned classroom questions and all.
More courses are coming in January, and CMU has already said distributed systems is one of the ones people are asking for next. For now, four is what you get, and honestly, four free, current, name-brand CS courses with a working prerequisite map is already more than most "free university course" announcements actually deliver.
New AI tools, automation workflows, and course drops — straight to your inbox. Join 2,400+ builders.
A researcher found Windows 11 checks for claude.exe, codex.exe, cursor.exe and copilot.exe and reports which ones you run. Here's how to check and disable it.
Qwen4-Max left preview on November 16 at $1.50/$4.50 per million tokens. Here's how its real benchmarks and coding tests stack up against Claude Opus 5.5.
Google Antigravity's /teamwork needs a paid plan and can flood your folder with files. Claude Code subagents are cheaper but need manual orchestration. Here's which wins for long projects.
Comments