Carnegie Mellon just put 13 of its computer science courses on a free YouTube schedule. The two CMU courses I'd tell any working developer to take first aren't on the list.
The new program lives at cs.cmu.edu/learn, tagline "Computer science from Carnegie Mellon. Free, online, at your pace." Eight courses premiere this month, five more in January 2027. The lineup is heavy on ML and software process, with nothing on databases or operating systems. If you're a self-taught dev picking which of the free CMU computer science courses on YouTube to watch first, start with two that were already free (15-445 Database Systems and 15-213 Computer Systems) and get to the new ones after that.
Below is the order I'd go in, plus which courses let you submit code and which are video only.
The 13 CMU courses going up on YouTube
From the CS Learn page, premiering November 2026:
- 10-605 Machine Learning with Large Datasets (William Cohen)
- 15-110 Principles of Computing (Franceska Xhakaj)
- 17-214 Principles of Software Construction (Hammad Ahmad), Nov 17
- 15-442 Machine Learning Systems (Tianqi Chen and Zhihao Jia), Nov 17
- 17-635 Software Architectures (Swarnalatha Ashok and Scott Pavetti), Nov 18
- 17-643 Quality Management (Jeffrey Gennari and Christopher Timperley), Nov 18
- 11-768 AI Agents (Daniel Fried and Graham Neubig), Nov 19
- 17-645 Machine Learning in Production (Christian Kästner and Bogdan Vasilescu), Nov 19
Coming January 2027: 15-259 Probability and Computing, 11-777 Multimodal Machine Learning, 17-619 Product Management Essentials, 07-280 AI and Machine Learning I, and 17-627 Requirements for Embedded Systems.
The page says "Free lectures now on YouTube. Free courseware coming next," and promises lessons with "instant feedback" and a badge, plus AI tutoring "coming to courseware." There are no dates for any of that. There's also no license listed. In the Hacker News thread one commenter put it as "being 'public on YouTube' is not the same as open," and nobody from CMU answered the license question.
Lectures without labs are just Netflix
The best comment in that thread wasn't about which course to pick. One commenter said watching someone explain something without doing it "just gives rise to the illusion of fluency, not actual competence." Another wrote: "Lectures are nice, but without the projects and some kind of way to get feedback on it, lectures can fall flat a bit."
I agree, and the plan below is built around it. A CMU course is worth roughly what its labs are worth. Ninety minutes of Andy Pavlo explaining B+ tree latch crabbing feels like learning, but you only learn it when you write the concurrent B+ tree yourself and an autograder fails you on a race condition.
So I sorted everything by one question: can someone outside CMU get feedback on their code?
| Course | Where | Assignments for outsiders? |
|---|---|---|
| 15-445 Database Systems | CMU-DB YouTube + course site | Yes. Public Gradescope with an entry code; autograder opens after the CMU due date |
| 15-213 Intro to Computer Systems | CS:APP labs | Yes. 11 labs with self-study handouts |
| 17-645 ML in Production | CS Learn + GitHub | Labs, assignments and past exams on GitHub under Creative Commons; no grader |
| 10-605 ML with Large Datasets | CS Learn + 10605.github.io | Being released. Staff say most include test cases |
| 15-442 ML Systems | CS Learn + mlsyscourse.org | Enrolled students submit via Autolab; I found no public grader |
| 17-214, 17-635, 17-643, 11-768, 15-110 | CS Learn | Video only as far as I can tell |
The 10-605 note comes from a course staff member in the HN thread: "Most of the assignments include test cases for you to check the correctness of your work." The commenter, Jacob Rast, is listed as the course's education associate. That's a change from the 10-605 site, which says unofficial auditors "will not be given access to course materials such as homework assignments and exams."
Which CMU courses to watch, in order: a 6-month plan
Assume 6 to 8 hours a week next to a full-time job. Watch at 1.5x. Spend most of the time on code.
Months 1 and 2: Systems and databases (the old free stuff)
Start with 15-445/645 Intro to Database Systems. Lectures go up on the CMU-DB YouTube channel, and the FAQ says it will "make the auto-grader for each assignment available to non-CMU students on Gradescope after their due date." The projects are built on BusTub:
git clone https://github.com/cmu-db/bustub.git
Two rules from the course: don't publish your solutions on GitHub, and don't file project questions as GitHub issues. There's an unofficial Discord, which the FAQ says isn't monitored by instructors or TAs.
Run the CS:APP labs from 15-213 in parallel: Data Lab, Bomb Lab, Cache Lab, Malloc Lab, Shell Lab, Proxy Lab. If you run Attack Lab, the self-study targets need the -q flag so they don't try to reach a grading server. Someone on HN said they'd love to try malloc again as an adult. The Malloc Lab handout has been sitting on that page for years.
These go first because every course after them assumes you know what a page, a cache line and a syscall cost. ML systems is mostly those three problems at a bigger scale.
Month 3: Software construction and architecture
Watch 17-214 Principles of Software Construction, then 17-635 Software Architectures. Both are video only, so do your own exercise. Take a service you maintain at work and redraw its architecture using the course's vocabulary. If you can't explain why a boundary sits where it does, you've found next week's refactor.
Skip 17-643 Quality Management unless QA process is your actual job.
Month 4: 15-442 Machine Learning Systems
This is the one I'd fight for. Tianqi Chen created TVM, so you get ML compiler material from someone who built one. The course site is the Spring 2026 offering. Follow the schedule there and treat the GPU and memory lectures as the database course again, with tensors in place of tuples.
Month 5: 10-605 and 17-645
10-605 covers ML at scale and has the test-case assignments. 17-645 Machine Learning in Production is the most practical course on the list for anyone shipping a model into a real product. Its materials are on GitHub under Creative Commons, and the textbook is free online.
Month 6: 11-768 AI Agents
I put Fried and Neubig's course last on purpose. Agent frameworks change every quarter, and after five months of systems work you'll judge an agent design on what doesn't change: latency, failure modes, evaluation. A HN commenter notes the YouTube playlist lags behind the most recent lectures, so check the date before you binge.
Put 15-259 Probability and Computing (Harchol-Balter, Wang, Saad) on your calendar for January 2027. Queueing theory is the closest thing backend engineers have to a superpower, and very few people learn it on purpose.
"Why learn this if the agent writes the code?"
This is the real fight in the HN thread, and people are split. One camp says low-level knowledge "goes in the bin" and you just re-prompt. The other asks the question I keep coming back to: how do you tell whether the output is correct without the fundamentals?
I'm with the second camp, and it's not close. An agent will happily write you a buffer pool with a lost-update bug. If you've never written one, you'll merge it. The developers who'll still be worth hiring are the ones who can read generated code and spot the lock that's held too long, the N+1 query, the O(n²) hiding inside a list comprehension. That's 15-445 and 15-213 material, and AI makes it more valuable because there's far more code to review now.
Where the "it goes in the bin" crowd has a point: you don't need all 13 courses. 15-110 is an intro course (and an HN commenter complained about its audio). The product and requirements courses are fine, but they won't change how you write code.
A checklist for this week
- Subscribe to the CS Learn YouTube channel from cs.cmu.edu/learn and fill in the "Keep me posted" form so you hear when the courseware with feedback ships.
- Clone BusTub and get it building.
- Download the CS:APP Data Lab self-study handout and finish it before you watch any ML lecture.
- Block two evenings a week on the calendar.
Which CMU course are you starting with? And what's the one CS fundamental you wish you'd learned earlier? Mine is concurrency. Tell me in the comments if you think the "fundamentals are the moat" argument is wrong now that agents write most of the code.
Top comments (0)