Something I wholeheartedly believe is that Google is winning the AI race. A lot of people would disagree with that, especially when tools like ChatGPT dominate the conversation, but I genuinely think Google is working toward something far more interesting (and more practical). And no, I'm not just talking about Google's AI chatbot itself, Gemini. While Gemini 3.0 is undeniably powerful, and it's clear why it shook the AI space like it did, what really stands out is the growing ecosystem of tools the tech giant is building around it.

A bunch of Google's Labs experiments (and many that are now widely available) are all powered by Gemini under the hood, and instead of just acting as standalone chatbots, they're designed to fit into your existing workflows. NotebookLM and Antigravity are both great examples of this approach, and using them together just proves my point. \

So, what are NotebookLM and Antigravity, anyway?

Google's best AI tools

Depending on the nature of your work, you might have heard of one tool but not necessarily the other. So, let's begin with a very quick rundown of what each tool is. NotebookLM is an AI-powered research assistant, and its superpower is that it grounds all its responses entirely in the sources you provide.

Instead of pulling information from the web or its training data, it lets you work directly with your information. Ultimately, this makes NotebookLM far more reliable than the vast majority of AI tools out there for research-heavy work where accuracy and context actually matter. On top of NotebookLM's superior RAG capabilities, it is also packed with features designed to make research easier and far more interactive, like Audio Overviews, Video Overviews, Mind Maps, Slide Decks, and more.

Antigravity, on the other hand, is a completely different beast. It's Google's AI-powered "agentic development platform" that the company quietly announced alongside the launch of Gemini 3.0. It's an IDE built on top of VS Code's open-source base, but with an interesting twist: it's built around AI agents (not just AI assistance).

This effectively distinguishes it from Cursor or VS Code (when using Copilot), where you're still the one writing code line by line (or, let's be honest, copying and pasting whatever the AI assistant spits out). Antigravity skips the copy-paste step entirely and lets you manage AI agents that can plan, build, run, and even test your project on their own.

Pretty much all the AI tools we've seen so far let you treat AI as an assistant. AI suggests, auto-completes the code, and helps you move from idea to execution much faster. However, at the end of the day, you're still the one making decisions. With Antigravity, you can switch to its Manager mode, which lets you deploy agents that handle every aspect of the development process autonomously.

You describe what you want to create, and the agents break it into tasks, write the implementation code, and test it themselves. All you do is sit back and review their progress — occasionally granting them permission to do something they're not sure about.

Antigravity is currently available in public preview and is completely free for individuals. The IDE is compatible with macOS, Windows, and Linux.

NotebookLM helps me come up with a plan

And actually understand it

I'm a computer science major, and while I've always enjoyed coding and problem-solving, I'm in no way skilled enough to build full-fledged projects from scratch. Instead of taking a lengthy class or reading tutorial after tutorial, I've always learned best from just diving into one of the many projects I have and just... beginning.

Before AI was what it is today, building something (even when you're in the learning phase) meant watching YouTube videos, reading through Stack Overflow threads, copying code snippets you didn't fully understand, and praying it all worked when you merged it together. Today, this process is a lot smoother. Going from an idea to a solution takes a detailed prompt, and tools like Antigravity are designed for exactly that. So that's what I tried to do first.

But here's what I've learned: the quality of what you build is only as good as your understanding of the problem. For instance, something I noticed recently is that my classmates and I aren't able to figure out our grades until our professors release them (even when we have all our marks). This is because our college recently made the switch to curve-based grading, and we don't quite understand how it works. Without knowing the exact formula they use, we're just guessing until the results come out.

Now, given that I wanted to build a tool to solve this problem, I had a pretty obvious issue: I didn't understand the formula either. Here's where NotebookLM comes in. I uploaded a bunch of credible sources that highlighted how this grading system works, and I also dropped in a result sheet my professor had shared with the class. It had the raw boundaries (A starts at 68, B+ starts at 63, and so on) and how many students fell into each bucket. I figured if NotebookLM could analyze that alongside the theory, it could reverse-engineer the exact formula. As expected, NotebookLM explained the system to me perfectly, specific to the logic my college was likely using.

Now, here's the fun part — I realized that just as NotebookLM had explained the entire logic to me, it could also explain it to the AI agent. I could do it too, but given NotebookLM already had a bunch of sources and I had just asked it a lot of questions, it could structure that explanation into a format the agent would instantly understand.

I asked it to turn our conversation into a technical specification, effectively creating a Product Requirement Document (PRD) that I could feed directly into Antigravity. A PRD is essentially the blueprint and tells developers exactly what to build so they don't have to guess. If I had just gone to Antigravity and typed "Make me a grade calculator," it would've just given me a generic tool (which I'd have already been able to find online)

Antigravity handles the execution

And I just watched

Once I had dropped the PRD NotebookLM generated, I just sat and watched Antigravity build. At the end of it, it generated a walkthrough, which included steps I could use to test the tool, along with the features the grade calculator supported. Given that I had provided the IDE with an extremely specific prompt grounded in actual research, Antigravity didn't have to guess and simply built exactly what I had described.

I created another instance just to see what Antigravity would generate with a prompt I wrote myself. The result was nowhere near what I wanted, and I had to jump in and fix issues myself or ask the agents to clarify, rework, and correct the implementation — something I didn’t have to do at all with the PRD generated by NotebookLM.

NotebookLM helped me actually learn from what Antigravity built

No point in just copy-pasting

As I mentioned above, my goal with this project was to learn. Just not in the traditional textbook-style method we've all grown tired of. I learn better by building first and understanding later. NotebookLM is an excellent tool for learning new skills, including programming.

For instance, I used the tool to learn Swift and Java, and my colleague has used it to learn advanced Python tricks! I also have a NotebookLM notebook filled with HTML, JavaScript, and CSS resources (since it's part of one of my programming courses). The notebook has guides, tutorials, YouTube explanations, and more as sources, making it my go-to resource for web development questions.

So, to actually understand this project, I uploaded all three of the files Antigravity had generated and combined them with my existing web dev notebook. Now I could ask NotebookLM to explain the code using the same resources I'd been learning from all semester.

The best way to learn and build with AI

If I'm not actively incorporating AI into my learning setup, I know that I'm just setting myself to be outpaced by people who are. NotebookLM's already a massive part of my learning process otherwise, and I instantly knew I wouldn't get very far.