A few days ago, I watched an episode of Harvard Thinking while working that’s been ingrained in my mind ever since. The episode brought together three experts to answer a question every student and educator has probably wondered: what happens to learning when AI can do your homework in minutes? The conversation was both fascinating and uncomfortable, and it raised some great points.

As a full-time student who just so happens to write a lot about AI and productivity for a living, I have a front-row seat to how my peers actually use these tools. I’ve seen classmates paste entire assignments into ChatGPT and submit whatever comes back without so much as glancing at it. I’ve seen group projects where nobody could explain the work they turned in. I’ve seen people give presentations where the scripts are blatantly AI-generated, and it’s clear the first time they’re being read is right there in front of us.

So when the podcast’s guests argued that AI threatens to erode the fundamental skills education is supposed to build, I nodded along. However, I also think most people are sleeping on just how transformative AI can be for learning when you use it right. That's why I've spent the last few months perfecting my own approach by building an AI-powered learning stack.

NotebookLM is where it always begins

My home base for learning

The very first article I wrote about studying and AI was about NotebookLM’s learning features. Since then, I’ve written countless articles about the tool, and it’s still the first one I open whenever I want to learn something new. NotebookLM is an AI-powered research assistant by Google, and its core functionality allows you to build grounded workspaces where you control exactly what the AI knows. This means that every AI-generated insight you get within a notebook is grounded in sources you’ve uploaded to that notebook, whether that’s an article, a YouTube video, your own notes, and so on.

While the tool has many great features that make the learning process more engaging and interactive, this grounding aspect is what makes it so invaluable as the foundation of my stack. When studying, whether that’s for an exam or for your own learning, it’s especially important to know that the information you’re working with is accurate and traceable back to a real source. With NotebookLM, every response comes with a clickable citation right next to it that shows you exactly where the information was pulled from.

That said, my learning process always begins the same way now: I create a fresh notebook for whatever I want to learn. Even if I don’t have a lot of solid material to add to it, I create it anyway and build it up as I go. I then use NotebookLM’s chat to ask questions, explain the fundamentals to me, and break down concepts. I essentially treat it like a tutor.

I also extensively rely on a feature called Notes. As I learn, I jot down key takeaways, my own thoughts, and anything I want to revisit directly inside the notebook. This keeps everything in one place.

Beyond just interacting with the material through questions, the Audio Overview feature has been incredibly helpful for learning. It turns your sources into a podcast-style conversation between two AI hosts, and it’s hands-down the best way to passively keep learning. I listen to them while commuting, taking walks, doing chores, or just when I need a change of pace from staring at text.

Gistr and Longcut come in next

For when you need to go deeper on a single source

While NotebookLM is a great tool for seeing the bigger picture and understanding how different sources connect, it’s not my preferred tool when I want to deep-dive into individual sources. For this purpose, I lean on two different tools interchangeably: Longcut (formerly TLDW) and Gistr.

When you’re trying to summarize YouTube videos or cross-question them in NotebookLM, you get a text summary that, while useful, strips away the personality and purpose of the original video. Longcut is my go-to for YouTube videos and takes a different approach: instead of compressing the video into text, it identifies the highest-signal moments as highlight reels and lets you watch directly from the source. You still hear the creator’s voice, see their visuals, and pick up on the nuance that a written summary would lose.

What I love about the tool is that it lets you decide how the highlight reels are generated. What’s important in a video to me might not necessarily matter to you. Longcut accounts for this by letting you define custom topics, so the highlights it generates are specific to what you’re trying to learn. Similar to NotebookLM, you can also ask any questions you have about the video and get answers with clickable citations.

Gistr covers a lot of the same ground for YouTube videos as well. However, where it really shines is how it handles other formats. While NotebookLM lets you upload PDFs, you can’t actually view the original document within the tool. Gistr displays the PDF right alongside the chat panel, so you can read through the actual content while asking the AI questions about it. You can also highlight passages, annotate directly on the document, and even turn highlighted text into questions with a single click.

I find this especially important when I’m learning something that involves documentation, lecture slides, research papers, or anything visually dense. Between Longcut for video and Gistr for everything else, I have a way to go deep on any individual source.

Recall keeps the learning going

Browsing the web is now part of studying

The tools I’ve mentioned so far all require me to sit down and actively study. However, something I’ve realized is that learning doesn’t always happen that way. I spend a lot of time browsing the web for work, watching completely random videos, and reading articles that aren’t directly connected to what I’m learning in my free time. More often than not, though, I end up coming across something that just… clicks. A concept that connects to something I learned last week, or a perspective that reframes a topic I’ve been struggling with.

Before Recall, those moments just… disappeared. If this is the first time you’re hearing of Recall, it’s a knowledge management tool that merges Obsidian, NotebookLM, and Anki into one neat tool. It has a Chrome extension that lets you quickly clip anything you come across into your knowledge base. The moment you do, Recall automatically summarizes the content, categorizes it, and connects it to related material you’ve already saved. It also builds an Obsidian-like knowledge graph that maps how everything you’ve saved connects to each other. The more you add, the richer the graph gets.

While Recall might sound a lot like NotebookLM, the key difference is that everything connects to everything. NotebookLM’s notebooks are deliberately isolated, which is great for keeping context clean but means your learning stays fragmented across separate workspaces. Recall treats your entire knowledge base as one living, interconnected system.

Every learning stack still needs a great conversational AI

For when nothing else clicks

Finally, it’s 2026, and everyone uses some sort of AI chatbot. Whether it’s ChatGPT, Gemini, Microsoft Copilot, Perplexity, or Claude, most people have a go-to they turn to for quick answers. In the context of learning, though, I think most people massively underestimate just how useful a conversational AI can be when you treat it less like a search engine and more like a tutor.

I’ve always been the type of person who asks my professors a hundred questions, so AI has been a natural extension of how I already learn. My go-to right now has been Claude. When I’ve gone through a source in NotebookLM or Gistr and still can’t wrap my head around something, Claude is where I go to think out loud. I ask it to rephrase things, build analogies, challenge my understanding — simply whatever it takes to make a concept stick. It’s the most patient study partner I’ve ever had.

When used right, AI is the best study partner you'll ever have

Like I mentioned in the very beginning, I'm not in favor of using AI to cheat through education. That's not what this is about. Every tool in my stack is designed to help me engage with material more deeply. The difference between using AI to learn and using AI to cheat comes down to one thing: intention.