Stop Saving Tutorials You'll Never Watch. Forward Them to an AI That Learns Them For You.
Forward any AI tutorial video to a Telegram bot and an OpenClaw agent watches it, writes itself a skill file, and uses the method automatically on future tasks.
Tools
- OpenClaw: a free, open-source AI assistant that runs on your own computer and talks to you through messaging apps you already use. Think of it as an employee who lives in your chat.
- Telegram: the free messaging app that acts as the inbox here. You forward videos to your agent the same way you'd forward them to a friend.
- Claude: the AI model doing the actual watching, listening, and writing. OpenClaw is the body, this is the brain.
The key concept, no tool link needed: a skill file. It's a recipe card your AI writes for itself, in plain English, and pulls out when the job comes up again. That's it. Not code. A text file with instructions.
The Story
Everyone has the folder. Saved Instagram reels, bookmarked YouTube videos, a "watch later" list with 200 entries. Every one of them was saved with real intent: this trick could save me time, this could get me customers, I should set this up. And every one of them is dying in there, because saving something and acting on it are two different jobs, and you only ever do the first one.
A creator called Trial And AI had the same problem, with a twist: his friends and clients kept sending him AI tutorial reels, and he was the one losing them. So he stopped trying to be the student and made his AI agent the student instead.
Now when someone sends him a reel explaining some AI hack, he forwards it to his OpenClaw agent on Telegram. In his words: "That's it. My job is done there."
The agent takes it from there. It pulls the video apart three ways: it listens to the audio, reads the transcript, and looks at the actual video frames, because plenty of tutorials show the important steps on screen without ever saying them out loud. Once it has understood the method the video is teaching, it writes itself a skill file capturing that method as step-by-step instructions.
Then comes the part that makes this more than a note-taking trick. The next time he gives the agent a task where that method applies, the agent picks the skill up on its own and does the work that way. He doesn't have to remember the reel exists. The agent does.
There's even a quality gate: if the agent already has a skill for that kind of task, it only replaces it if the new method from the video is actually better than what it already knows.
His summary of the before and after: "Up till now, all my saved reels were dying in my folder. But now, every forwarded reel is basically a skill addition to my agent."
How It Works
The flow fits on a napkin:
FRIEND SENDS YOU A TUTORIAL REEL
|
v
You forward it to your agent on Telegram
|
v
Agent pulls the video apart:
audio + transcript + video frames
|
v
Agent writes a skill file
(plain-English recipe card: "when asked
to do X, here's the method")
|
+-- agent already knows a way to do X?
| keeps whichever method is better
|
v
WEEKS LATER: you ask for something where X applies
|
v
Agent finds the recipe card on its own
and does the job that way
The mechanics under the hood are refreshingly boring. OpenClaw agents keep a skills folder on your machine. Each skill is just a file, usually called SKILL.md, with a name, a one-line description of when to use it, and plain-English instructions. Before starting work, the agent scans those descriptions like an index. When a task matches one, it opens the full recipe card and follows it.
That's why this whole system is possible. Teaching the agent something new doesn't require training, programming, or a developer. It requires writing a clear text file into a folder. Which happens to be exactly the kind of thing an AI agent is good at doing for itself.
So the "learn from a video" trick is really just a chain of three abilities the agent already has: receive a file on Telegram, analyze audio and images, and write text files. The builder's insight was pointing those three abilities at his saved-posts problem.
Your Saved Folder Is a Liability. This Makes It an Asset.
Here's the replacement math, because there always is some.
Think about what actually happens to a good tutorial today. Option one: you watch it, get inspired, and do nothing, which is free but worthless. Option two: you watch it, spend three hours implementing it yourself, once, and forget the details within a month. Option three: you pay someone, a VA at $500 a month or a freelancer at $50 an hour, to "set up that thing from the video," and now the knowledge lives in their head, not your business.
This flips all three. The knowledge lands in a file, in your folder, on your machine, in a form your AI acts on automatically. Forward ten good tutorials over a month and your agent has ten new standing capabilities. Nobody watched anything. Nobody billed you.
The compounding is the real prize. A saved reel is worth the same in a year as it is today: nothing. A skill file gets used every time a matching task shows up, forever. One is a bookmark. The other is an employee who took the training course for you and actually retained it.
The Honest Caveats
Two things the one-minute video glosses over, and you shouldn't.
First: garbage tutorial in, garbage skill out. The agent learns the method as described. It does not fact-check the influencer. A huge share of AI tutorial reels are recycled, outdated, or flat-out wrong, and if you forward one of those, your agent will faithfully write itself a recipe card for doing something badly. You are the editor of your agent's curriculum. Forward the stuff from people who clearly run the thing they're teaching, skip the "THIS AI TRICK BROKE THE INTERNET" genre.
Second: read what it wrote, at least early on. A skill file is plain English, so this takes two minutes. Open the file, check that the steps match what the video actually showed, and fix anything it garbled. After the first handful you'll know how much you can trust the pipeline. The "only replace an existing skill if the new one is better" rule is also a judgment call the agent makes, so spot-check that too before you let it overwrite methods you rely on.
Neither caveat kills the idea. They just mean the system is a junior employee, not a magic box. Junior employees need their work reviewed for the first few weeks. Then they mostly don't.
Who Should Steal This
Anyone whose saved-posts folder has more than twenty entries, which is everyone reading this. More specifically: the owner who keeps getting "have you seen this AI thing?" texts from friends and feels a small stab of guilt each time. The agency or consultancy operator whose whole edge is staying current on tactics, because this turns staying current from a reading habit into an automatic pipeline. And any small business owner already running an AI agent, because this is the cheapest upgrade available: your agent stops being a tool you have to teach by hand and starts being one that takes the class for you.
The reel your friend sends you tonight can either die in the folder with the other 200, or become something your business knows how to do. Same thirty seconds of effort either way.
Keep Reading
- Teach Your AI to Stop Making the Same Mistake Twice: the sibling pattern, where the agent writes skill files from its own failures instead of other people's tutorials.
- Your AI Remembers Everything So You Don't Have To: the memory layer that makes a learning agent worth having.
- Give Your AI a Second Brain That Gets Smarter Every Day ($0/Month): the plain-text-files philosophy behind all of this, taken to its logical end.