I finally admitted something to myself: of all the courses I bought on Coursera over the past few years, fewer than one in ten were ever completed. Yet a random interview I clicked on YouTube one evening led me to revise my agent prompts across three versions that same night.
Coursera Sold Me a Illusion, YouTube Actually Fed Me
Coursera’s course pages are designed to look like learning: chapters, quizzes, certificates, progress bars. But when you actually sit down to watch them, your attention drifts. I’ve figured out why since — those courses are packaged knowledge, and the packaging itself is a tax. You have to move at its pace, follow its sequence, and by week three you’ll finally stumble onto the one concept you actually wanted from the start.
YouTube is the exact opposite. It’s the largest learning platform in the world, yet it looks nothing like a learning platform. Search for a concept and one of the top three results will someone break down an entire textbook chapter in twenty minutes. If they miss something, the comments will fill in the gaps. There’s no progress bar holding you hostage, no one tracking your completion rate. There’s also a lot of junk, but the upside of junk is that the cost of filtering is near zero — click for thirty seconds, realize it’s not useful, and scroll past.
Text Gives You a Map, Video Gives You a Journey
When I learn something myself, my first choice is always a text-and-image tutorial. The reason is very specific: I read quickly. With a text tutorial, I can skim the table of contents and headings first, get a full overview of the knowledge landscape, and then decide which sections to read in depth. That’s a map. Video can’t do that. A video is a journey — you can only follow along with the camera, and you can’t see the entire structure in the first second.
So when I’m looking up a concept, doing a specific task, or following a tutorial on my own, text is always faster. It’s pure information-acquisition arithmetic.
Interviews Are Something Else Entirely
But recently I’ve realized there’s a scenario that neither text nor tutorials can cover: watching interviews.
When I watch an interview with someone truly impressive, I’m doing three things simultaneously:
- When I encounter a concept I don’t know, I look it up on the spot, then come back and keep watching;
- I take notes on the ideas they’re sharing;
- I use those ideas in real time to write and iterate on my own agent prompts.
The third thing is the key. Interviews don’t give you the “step one, step two” structure of a tutorial. They give you the interviewee’s way of judging and making trade-offs. You can’t capture that kind of thing through a text summary. You have to watch the person actually speak — see where they pause on a certain topic, what they dismiss with a wave of their hand — to actually fish it out.
And the most effective way to use what you’ve fished out is to write it into your system immediately. Watch a segment, switch away and revise a prompt, then switch back and keep watching. The interview becomes a real-time, feedback-driven learning loop: ideas go in, system behavior comes out, and the changes are visible.
That feeling is infinitely more satisfying than finishing an entire course. Because the output of a course is “completion,” while the output of this loop is my system actually getting better.
It’s Not About the Tool, It’s About the Loop
Looking back, the difference between Coursera and YouTube isn’t content quality — both have great content. The difference is that one pins you to your seat as a student, while the other lets you jump in and act like an engineer at any moment.
When it comes to learning, the best state is always the latter: you have something you’re actively improving, and every piece of input gets digested on the spot. Interviews, text, and video are all just input channels. The loop is what’s yours.
Tonight, after three rounds of prompt revisions, my agent’s test pass rate went up compared to yesterday. That feels more like actual learning than any certificate of completion ever could.