My YouTube Watchlist Just Got an AI That Watches for Me#

Wake up to a clean summary of every video you care about — no more 2x speed marathons.

The notification lie#

You subscribe to a channel. You click the bell. You check "all notifications."

Then the video drops. And it never appears in your feed. Not in notifications. Not on your homepage. It just vanishes into YouTube's recommendation black hole.

This isn't a you problem. It's a YouTube problem. Their algorithm optimizes for watch time, not for showing you the creators you explicitly followed.

So you end up with two bad options: manually check each channel like it's 2009, or accept that you're missing half the content you actually want to see.

I went with a third option. I made an AI do it.

The setup that actually works#

There's a skill called youtube-full that handles the entire pipeline — fetching new uploads, pulling transcripts, summarizing, and delivering a digest. You install it once and tell it what to track.

npx clawhub@latest install youtube-full

Or just tell your OpenClaw: "Install the youtube-full skill and set it up for me."

The agent handles account creation, API key setup, everything. You get 100 free credits on signup. No credit card.

After that, the key gets stored in the right location for your OS automatically. No env file wrangling.

Why not just use yt-dlp?#

I tried that first. Bad time.

The youtube-full skill uses TranscriptAPI.com under the hood. When you're chaining API calls through an agent, the last thing you want is yt-dlp dumping 50 lines of debug output into your context window. Clean JSON means fewer tokens wasted, fewer parsing errors, fewer "why did this break at 3am" moments.

Two ways to run it#

Option 1: Channel tracking#

I give it a list of handles and tell it to check every morning:

Every morning at 8am, fetch the latest videos from these YouTube channels and give me a digest with key insights from each:

- @TED
- @Fireship
- @ThePrimeTimeagen
- @lexfridman

For each new video (uploaded in the last 24-48 hours):
1. Get the transcript
2. Summarize the main points in 2-3 bullets
3. Include the video title, channel name, and link

If a channel handle doesn't resolve, search for it and find the correct one.
Save my channel list to memory so I can add/remove channels later.

If a handle is wrong or changed, the agent searches for the right one. No manual ID hunting.

Option 2: Keyword tracking#

Sometimes I care more about a topic than a specific channel. For that, there's a deduplication pattern using a simple text file:

Every day, search YouTube for new videos about "AI agents".

Maintain a file called seen-videos.txt with video IDs you've already processed.
Only fetch transcripts for videos NOT in that file.
After processing, add the video ID to seen-videos.txt.

For each new video:
1. Get the transcript
2. Give me a 3-bullet summary
3. Note anything relevant to my work

Run this every morning at 9am.

This keeps you from burning credits on the same video twice. Simple, effective, no database required.

The credit math matters#

Here's what I like about the pricing structure:

• channel/latest and channel/resolve calls are free — zero credits
• Only transcript fetches cost 1 credit each
• 100 free credits on signup means you can process 100 videos before paying anything

So checking whether channels have new uploads is unlimited. You only spend when there's actually new content worth summarizing. That's the right incentive alignment.

You can customize the output#

The digest format isn't locked. You can ask for whatever format fits your workflow:

• Key takeaways (default, 2-3 bullets)
• Notable quotes pulled from the transcript
• Timestamps of interesting moments so you can jump straight to the relevant section
• A "relevance to my work" assessment for research-heavy workflows

I've been running the key takeaways version for two weeks. My morning routine went from "scroll feed for 20 minutes, watch nothing" to "read digest for 3 minutes, watch 1-2 videos that actually matter."

The catch#

This isn't a novel idea. Recapio already sells daily YouTube recaps as a product. The difference here is control — you pick the channels, you pick the format, you pick the schedule, and the agent adapts without you clicking through a SaaS dashboard.

If you want something that works out of the box with zero config, Recapio is fine. If you want something that fits your workflow and evolves when you tell it to, the agent approach wins.

If you're tired of YouTube deciding what you see, set up the youtube-full skill and take your feed back. Check out papayaclaw.com for more agent workflows that handle the stuff you shouldn't have to.