By  Mahnoor Faisal Published Jul 7, 2026, 6:00 AM EDT Mahnoor Faisal is a tech journalist covering AI and productivity tools with bylines at XDA, SlashGear, MakeUseOf, Laptop Mag, and Android Police.

She’s been writing professionally since she was sixteen, and has since penned hundreds of articles. This includes in-depth coverage of AI tools like NotebookLM to breaking news across the AI space. Her passion for technology started when she received her first iPod Touch (4th generation) on her 8th birthday, and she’s been deep in the tech world ever since.

Currently pursuing a degree in computer science, Mahnoor brings both a journalist’s eye and a technical foundation to her coverage of how AI is reshaping the way we work and learn.

By Mahnoor Faisal, Published Jul 7, 2026, 6:00 AM EDT

There aren’t a lot of AI tools I speak highly of so often, and NotebookLM is one of the few exceptions. From day one, Google’s NotebookLM has impressed me with the concept it’s been built around: grounding everything in your own sources instead of pulling answers out of thin air. Unlike a lot of other tools that end at impressing me once and losing their entire charm, the team has kept finding new ways to earn praise. Video Overviews, Slide Decks, Mind Maps, Infographics, and the list goes on and on.

While I’ve mostly had only good things to say about the tool, the recent update is one of the few times where I’ve been very vocal about my disappointment with it. NotebookLM seems like it’s not NotebookLM anymore in many ways, and I keep feeling like Google’s trying to turn it into Gemini (or GeminiBookLM, as I like to term it). However, one of the few parts of this update I can’t fault is code execution. After testing it out, I came away genuinely impressed, to the extent that I handed it a task I’d normally do myself and watched it, not just keep up, but to catch something I’d have missed.

Yes, NotebookLM can now write and run code

GeminiBookLM, but make it useful

Given that I had early access to this particular NotebookLM update, I remember being so confused as to why Google felt the need to add code execution to a tool they market as a research assistant. Think about it, it makes no sense. The entire point of NotebookLM has always been to read your sources and answer from them, not to open a terminal and start running scripts. Code execution felt like a solution in search of a problem, and the kind of feature that gets added because a model can do it rather than because anyone asked.

As of the June 8 update, each notebook is now equipped with a secure cloud computer, enabling NotebookLM to write and run code for deeper research and more complex analysis. There’s now an isolated, sandboxed environment tied to every notebook where the tool writes scripts (usually Python), actually runs them, and hands you back the result. That’s what shifts it from describing a number to computing one.

Powering all this is a change under the hood. NotebookLM now runs on Gemini 3.5 and Antigravity, providing more accurate and reliable information along with better visibility into the thinking process. The cloud computer is where code executes and Gemini 3.5 plus Antigravity are what let it plan and drive that execution. On top of the VM sit more than 100 curated software skills, which are pre-built capabilities that the tool reaches for mid-task by figuring out which ones the job needs.

Google says the payoff shows up in its own benchmarks: an average win rate of over 65% against the prior system across its top five evaluation dimensions, including 69.9% on large document analysis and 78.2% on advanced web research and source discovery.

Related

https://www.xda-developers.com/engineer-behind-notebooklms-best-feature-showed-me-his-exact-setup

The engineer behind NotebookLM’s best features showed me his exact setup, and it changes everything

198 articles later, I finally got to ask the team how they actually use it.

The results were far better than they had any right to be

My skepticism didn’t survive the first test

I’ve used NotebookLM since its early Google Labs days, and I know well what it’s good at and where it tends to fall short. I’ve tried to use it for programming (I don’t know why either), and it wasn’t great. I’ve also used it for a lot of data analysis before, and the results weren’t all that great either.

So I began my testing with fairly low expectations. To begin my testing, I gave it something real. My semester just ended, and something I’ve turned to a bunch of LLMs for is figuring out how my grades might land before the official results come out by feeding in a class result sheet and asking what the curve might look like. It’s the kind of task that sounds simple but isn’t.

Read more: I used NotebookLM’s code execution and I wasn’t ready for how well it went – XDA

Continue/Read Original Article: I used NotebookLM’s code execution and I wasn’t ready for how well it went