Tutorial hell got a sequel: vibe-coding hell. How to learn from coding videos in 2026

For years the warning for new programmers was tutorial hell: you watch video after video, everything makes sense while the instructor types, and then you open a blank editor and can't build anything. In February 2026, Lane Wagner of Boot.dev described its sequel, vibe coding hell: learners who can build projects, because an AI writes most of the code, but don't understand how those projects work.

They look like opposites. One person can't build anything and the other builds constantly. But they share a cause, and the same fix works for both. I build Unwatch, a tool that turns coding videos into written lessons, so I'll say where a tool helps and where it doesn't.

What the two traps have in common

In both cases, someone else does the hard part and you feel the understanding. Watching an instructor solve a problem is fluent and satisfying, and so is reading an AI's answer that works. Neither makes you retrieve anything from memory, struggle with a bug, or explain a design choice. The feeling of understanding comes without the learning. Research on learning has pointed this way for a long time: attempting to recall or produce something beats re-reading or re-watching it (for example, Roediger and Karpicke's 2006 paper on test-enhanced learning).

Wagner's advice for the AI version is to turn off autocomplete and agent tools while you're learning, and to use a chatbot for explanations and Socratic questions rather than for writing your code. The video version has the same rule: don't let the tutorial do your thinking.

A loop for learning from a coding video in 2026

1. Predict before you watch. Read the title and description, then write down how you'd build it: the pieces, the data, the first function. It takes five minutes and the video now answers a question you actually have.
2. Skim, then watch only the hard parts. You don't need every minute of a 90-minute video. A transcript or written version lets you see the structure and find the sections you don't already know. Then watch those parts at normal speed.
3. Close it and build. Write the code without looking, even if it's wrong. The struggle is where learning happens, and the mistakes show you what you didn't understand.
4. Compare, then fix your version. Only now look at the video's code. Where it differs from yours, work out why, and don't just copy the lines.
5. Explain it back. Write three sentences on how it works, as if for a friend. If you can't, you have found the gap.
6. Rebuild tomorrow from a blank file. Then again in a week. This is the part almost everyone skips, and it's the part that sticks.

Using AI as a tutor without falling into the second trap

• Ask for a hint, not the solution.
• After you paste in code that works, ask the model to quiz you on why it works, and answer before it tells you.
• Ask it to find the bug in your code, not to rewrite your code.
• Check anything important against the documentation or the original video. AI makes confident mistakes.

Where a tool helps, with a test

The video loop above is slow if the code is only on screen. You can't search it, copy it, or compare it quickly. A written lesson with timestamps helps with steps 2 and 4: you see the structure, jump to the exact minute, and compare your code with the instructor's.

It matters that the code is exact. I ran one 25-minute Python video through three summarisers. None reproduced the setup lines exactly as shown on screen, and two wrote html.parser where the screen shows 'html'. It was one video and one run each, and you can see the details in the comparison. A lesson that reads the screen avoided that, and you can see it in the sample lesson. Even so, it's AI-made, so compare it with the video before you rely on it.

A tool does not do steps 1, 3, 5 or 6. Those are yours, and they're the point.

The short version

Tutorial hell is watching without building. Vibe-coding hell is building without understanding. The fix for both is the same: do the hard part yourself, then use videos and AI to check your work. If you try the loop, I'd like to hear which step you found hardest.