I Learn Faster Than 99% of People. NotebookLM + Claude Code + Obsidian

ELI5/TLDR

A guy noticed that most people watch endless self-improvement videos and then change nothing. So he wired three tools together — Google’s NotebookLM (a chatbot that only answers from sources you feed it), Claude Code (an AI assistant that lives in a terminal and can do things on your computer), and Obsidian (a note-taking app) — into an assembly line. One end eats 200 Huberman podcasts. The other end spits out a morning checklist that asks him if he got sunlight today. The point isn’t the podcasts. The point is closing the gap between “I heard something smart” and “I actually did something about it.”

The Full Story

The video opens with a claim that sounds like it was written by an ad targeting you at 2 a.m.: ninety percent of people listen to experts and never change a single behavior. Leaving aside whether that number comes from anywhere, the underlying observation is hard to argue with. You watch a podcast, you nod, you close the tab, you do nothing. A week later you can’t even remember what the podcast was about. Artem Zhutov’s pitch is that the bottleneck isn’t willpower. It’s that nobody has a system that takes an insight from a podcast and walks it all the way to “this is on your calendar tomorrow at 7 a.m.”

So he built one. It has three parts, and each part does exactly one thing.

Part one: NotebookLM as the expert on tap

NotebookLM is a Google product that does something narrower than ChatGPT on purpose. You give it a bunch of sources — PDFs, websites, YouTube videos — and then you chat with it. The important trick is that it will only answer from those sources, and when it answers it cites which source said what. Think of it as a chatbot that has read exactly the books you handed it, and nothing else, and has to show its work.

The problem is loading stuff in. NotebookLM does not let you say “import everything from this YouTube channel.” You have to paste videos in one by one. If your chosen expert is Andrew Huberman, the Stanford neuroscientist who hosts a very long podcast about sleep, sunlight, and cold plunges, that’s hundreds of individual links. Nobody is doing this by hand on a Saturday afternoon.

Part two: Claude Code as the hands

This is where the second tool comes in. Claude Code is Anthropic’s AI assistant that lives in a terminal — basically a chat window where the AI is allowed to run commands on your computer. You tell it what you want, it figures out the steps, and it does them. In the demo, Zhutov has rigged Claude Code up to drive NotebookLM on his behalf. He tells it, roughly: “go find Huberman’s channel, pull the last 200 health videos, make a new notebook, and dump them in.” And then he watches the counter tick up — 60 sources, 96, 200 — as Claude quietly handles the tedious part.

He also mentions “skills,” which in Claude Code are little reusable instruction packets. Once you’ve built a skill for “import a YouTube channel into NotebookLM,” you don’t have to explain it again next week when you want to do the same thing with Lenny’s Podcast, which covers product management. Same machine, different expert.

“We just turned research into experiments in just one session.”

So by the halfway mark of the video we have a searchable, citation-backed library of one expert’s entire body of work on a topic. That’s the “learning” half. Zhutov’s point is that almost everyone stops here, and stopping here is what makes self-improvement podcasts so inert. You have access to everything Huberman has said about sleep. So what. Nothing happens on Tuesday morning.

Part three: Obsidian as the life

Obsidian is a note-taking app where every note is a plain text file on your own computer. Dorky, beloved by people who still miss the early internet, infinitely customizable. Zhutov uses it as the place where his actual life lives — goals, a daily note, an experiment tracker, a health dashboard. This is what Claude Code writes to when it’s time to act on something.

Here’s the loop in motion. He tells Claude Code: my goal is better health, interview me using questions drawn from those 200 Huberman episodes, assess where I’m at, and propose experiments. Claude Code goes off and pulls questions out of NotebookLM in parallel — six at once, on supplements, exercise, sleep, and so on — saves the answers with their citations into Obsidian, and then runs him through an interview. He answers. Claude looks up his existing gym logs (which are already in Obsidian, because that’s where he tracks things), compares current state to target state, flags where the gaps are biggest, and proposes the three highest-leverage experiments.

In this run, the three are: sleep regularity, morning sunlight, and adding zone-two cardio. (Zone two is Huberman-speak for the moderate aerobic zone — the pace where you can still hold a conversation but wouldn’t want to sing.) Each experiment gets a hypothesis, a protocol, and a success metric. The sleep one, for example: hit your wake-up time within 30 minutes on 80 percent of days. This is the scientific method dressed in workout clothes.

“We are using scientific method and the success is okay, 80% of the days within 30 minutes of target wake up time.”

The experiments get written into Obsidian as files tagged “in progress.” Then the final piece: Zhutov has another Claude Code skill called “daily” — his morning routine. When he runs it, Claude reads his active experiments, asks him how each one went yesterday, writes his answers into the daily note, and helps him schedule today’s version. Did you get sunlight? How’s your mood on a scale of one to ten? Gym today or rest day? The questions aren’t fixed — they update whenever the experiments update.

That’s the loop closed. Podcast to insight to experiment to calendar to tracked data to review to adjustment. All of it in plain text files he owns, with every claim traceable back to a Huberman timestamp if he ever wants to double-check.

The meta-point at the end is that the specific expert is swappable. Swap Huberman for Lenny’s Podcast and you have a product-management assistant. Swap it for a chess coach’s channel and you have something else. NotebookLM is “external knowledge.” Claude Code is “the thing that does the stuff.” Obsidian is “the record of your actual life.” The three keep their jobs and you change the inputs.

Claude’s Take

The diagnosis is sharper than the demo. “People consume expert content and change nothing” is genuinely true and it’s the right problem to point at — the gap between knowing and doing is where most self-improvement content quietly dies. Giving that gap a name and building infrastructure for it is a real contribution, even if the infrastructure is mostly duct tape.

That said, a few things to hold at arm’s length. The workflow is only as good as the source, and Huberman in particular is a load-bearing choice that deserves a footnote. He is a real Stanford neuroscientist, but his podcast has been flagged repeatedly — by other researchers, by journalists — for overstating the certainty of animal studies, for leaning on supplement recommendations that outrun the evidence, and for presenting protocols with more confidence than the literature supports. NotebookLM’s citations solve one problem (did the expert actually say this) and do nothing about a bigger one (was the expert right). A “verifiable knowledge base” that verifies only that Huberman said it is still trusting Huberman. That’s fine if you know you’re doing it. It’s a trap if you mistake citations for evidence.

The n-of-one self-experimentation piece has the same shape. Running structured experiments on yourself is better than doing nothing, and tracking a mood score daily will teach you something about yourself. But “my energy is 6.8 on gym days and 6.1 on rest days” is not a causal claim, it’s a mood journal with a spreadsheet. The scientific-method framing is a little oversold — a real experiment needs controls, blinding, and enough data to rule out that you just felt good on Tuesday. Treat it as useful self-monitoring, not proof.

Third: the demo is a demo. Watching Claude Code smoothly drive a NotebookLM import is doing a lot of persuasive work here, and anyone who has spent time with AI agents knows they are also capable of producing elaborate, confident nonsense in between the moments that look clean on camera. The 20-minute setup claim is probably technically achievable for someone already comfortable in a terminal, and probably a full weekend for anyone else. And the whole stack — NotebookLM, Claude Code, Obsidian, plus the “skills” glue — is a pile of moving parts, each with its own bugs and billing, and each of which can quietly break in a way that makes your morning routine silently stop happening.

The actual transferable idea, stripped of the specific tools, is worth keeping: pair a searchable source of expert knowledge with an assistant that can write to the places your life actually lives (your calendar, your notes, your daily checklist), and you start to close the loop between reading about habits and having them. That’s real. Whether you need this exact stack to do it is another question. A piece of paper and a willingness to answer three questions every morning would cover maybe 70 percent of the benefit, and would never have a software update break it.