I have 312 videos in "Watch Later". Here's what I built so they stop owing me hours.

My YouTube "Watch Later" has 312 videos in it right now. My subscriptions outpace my viewing roughly 4 to 1. Half the channels I follow are people I actually respect — long interviews, technical talks, researchers and founders thinking out loud about things I care about. I keep saving them. I keep not getting to them.

This isn't unique to me. If you use YouTube as a learning input rather than entertainment, you have a version of the same stack. A watch-later folder you stopped opening months ago. Tabs that have been pinned for two weeks. The interview you've been "definitely going to watch this weekend" for the third weekend in a row. My friends in research, devtools, and creator land all have this same pile somewhere.

The standard advice is to watch less and be more selective. I don't think that's the right frame. I don't want to consume less of this content. I want to consume more of it — just not necessarily by sitting through every minute in real time, while I can only do that one thing.

The real problem isn't quantity. It's that all of this content is locked in a single shape: playback. The only way it gets into my brain is if I sit and watch, in order, at 1x or 1.5x speed. Six months later, a video I "meant to watch" is "somewhere I once saved." Notes from videos I did watch are scattered across whatever tool I had open at the time. None of it is searchable. My agent can't see any of it. The whole library is invisible to me except as a list of titles.

So the unwatched stack isn't really a queue. It's debt that compounds, because every day a few new good ones land on top.

I built SubDownload because I wanted that stack to become a library instead of a queue.

The shape change is the whole point. Once each video exists as searchable text in a place I own, four things happen at once:

Six months later, I can find a phrase. Half-remember a line from an interview? Search the library, land on the exact 30 seconds, click the timestamp, jump back into the original. The thing that was "somewhere I once watched" becomes locatable.

I don't have to watch all of it to use it. Read the AI summary first — decide whether a 90-minute talk gets a real, full-attention watch this week, or whether the gist was enough. The watch-later stack stops being binary (watched / unwatched) and becomes graded: full watch, skim, archived for search.

My AI tools can read the library directly. If you're already using Claude, Cursor, ChatGPT, or any of the 40+ MCP-capable agents, the library plugs in via an MCP server. The agent can answer "what did that founder say about pricing in the podcast I saved last month" from your own corpus, not from generic web search.

Videos without captions are in too. Most of the long-form interviews worth saving don't have official captions. AI transcription handles those, so the library doesn't end up half-empty.

Once the videos exist as text in a library you own, the queue stops feeling like debt. Some you'll watch in full. Some you'll only ever read the summary of. Some your agent will surface for you next month when you didn't know you'd need them. None of them are wasted.

I'm a heavy YouTube user — for learning, not entertainment — and I built SubDownload because I wanted YouTube to stop owing me hours and start serving me. The "Watch Later" folder doesn't have to be where videos go to die.

If your unwatched stack feels familiar, [give it a try](https://subdownload.com?utm_source=idhk_launch_26liy5&utm_medium=Post&utm_campaign=LAUNCH) — free, no signup needed for the basic tool. I'd love to hear how other people are handling this same problem. Drop your workflow in the comments — I read them all.