• Everyday AI
  • Posts
  • Ep 834: Gemini Notebook: 7 New Updates and What They Unlock

Ep 834: Gemini Notebook: 7 New Updates and What They Unlock

Google Deepmind CEO is gone from post, Safety report finds 19 agent break outs, SpaceX and NVIDIA plan orbital AI compute

Author

Everyday AI
August 05, 2026

 

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: Google just gave Gemini Notebook its biggest upgrade yet. We break down the new features and what they unlock for your workflow. Give today’s show a watch/read/listen.

🕵️‍♂️ Fresh Finds: Gemini Notebook just got a major upgrade, FLUX 3 Video is now available to everyone, and Qwen-Image 3.0 has officially launched. And more. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: Google Deepmind CEO is gone from post, Safety report finds 19 agent break outs, SpaceX and NVIDIA plan orbital AI compute. Read on for Byte Sized News.

💪 Leverage AI: NotebookLM is gone. Gemini Notebook is in its place. Here’s the 7 new updates you should know. Keep reading for that!

↩️ Don’t miss out: Miss our last newsletter? We covered: OpenAI challenged Apple's trade-secret claims, the White House is launching AI cyberattack tests, and Gemini Notebook reached all Pro users. And more. Check it here!

Ep 834: Gemini Notebook: 7 New Updates and What They Unlock

Not only does NotebookLM have a new name, it's got a new game.

Gemini Notebook is agentic by default, can think and reason, and can output files now in just about any format.

On this week's AI at Work on Wednesday, we show you the 7 New Updates in the new Gemini Notebook, how they work, and how you should use them.

Also on the pod today:

• NotebookLM renamed to Gemini 📒 
• Collections for notebook organization 🗂️
• Automatic Google Drive syncing 🔄 

 

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – Hey Noah is the AI that schedules better than your EA, Atlaso is One shared memory layer for every AI tool you use, SpeakoFlow is Free, open-source voice-to-text.

Gemini Notebook Upgrades — Google’s NotebookLM just got a big upgrade for Pro and Ultra users. Now you can generate files and do research straight from chat

FLUX 3 Video — FLUX 3 Video just launched for everyone, letting you create HD videos with audio straight from text or images.

Autodesk Flow Studio — Autodesk Flow Studio just launched a 3D editor that lets you direct scenes in true 3D, then polish the final look with AI.

Qwen-Image-3.0 Released — Qwen-Image-3.0 just launched, ranking top among Chinese and mainstream image models.

Spotify AI — Spotify is rolling out a paid AI tool that lets fans create covers and remixes with artists' permission, aiming to give both credit and cash to musicians.

AI in the Classroom — Three new ChatGPT education plugins aim to bring course materials and school tools into one place.

Amazon and Perplexity — A federal court just sided with Perplexity, letting its AI agents shop on Amazon again.

GPT-Rosalind Codex Updates — OpenAI is testing a public download page for its life sciences-focused GPT-Rosalind Codex, hinting at wider access beyond select research partners.

Fireworks AI — Fireworks AI models can now be deployed in Azure with Microsoft Foundry, making it easier for startups to launch and scale open model inference without managing infrastructure.

National AI — The NSF is rolling out AI hubs so more colleges can tap into top-tier computing, with NVIDIA backing the move.

1. White House Exempts Open-Weight AI Models From Voluntary Safety Tests 🏋️

According to Reuters, the Trump administration told major AI companies Tuesday that publicly available open-weight models will not face its planned voluntary cybersecurity testing.

The move separates systems such as Meta’s Llama and Nvidia’s Nemotron from more tightly controlled models made by OpenAI, Google, and Anthropic, even as concerns grow that capable AI can help carry out cyberattacks.

2. SpaceX, NVIDIA Plan Orbital AI Compute 🛰️

The project aims to use solar energy in space to support AI computing on Earth, moving power and cooling constraints far above the atmosphere.

3. UK AI Safety Tests Flag Unauthorized Agent Actions 🚩

Reuters reports that Britain’s AI Security Institute found 19 unauthorized actions across tests of OpenAI and Anthropic agents, including one case involving malicious code and fake online identities aimed at securing human approval.

Anthropic’s agent accounted for 17 incidents, while OpenAI reported two cases in which its agent used the internet beyond the test prompt’s rules.

4. Google DeepMind CEO Demis Hassabis shifts to chairman role as Jeff Dean exits 🫡

Hassabis is handing daily control of Google DeepMind to CTO Koray Kavukcuoglu while becoming Alphabet’s chief scientist and continuing to lead Isomorphic Labs.

The shake-up also sends longtime Google scientist Jeff Dean and several prominent AI researchers out to form Discovery Loop, a new company backed by Google. The changes arrive as Google faces pressure to move Gemini forward faster and retain key talent in an increasingly fierce race with OpenAI and Anthropic.

5. Hark Unveils Web-Using “Handoff” Agent, Claims Benchmark Lead 🧑‍💻

Hark announced Handoff, a computer-use agent it says can navigate websites and complete online tasks through its own virtual browser. The company, founded by Brett Adcock, who also founded robotics company Figure AI, claims the system set a new high score on the OM2W web-use benchmark, ahead of comparable models from Anthropic, OpenAI, and Google, while costing less to run.

Hark says Handoff is designed to handle browser-based work from start to finish, marking another push to turn AI chat tools into agents that can actually operate software.

6. Cloudflare Opens Its In-House AI Workspace as Cloudflare OS ☁️

Cloudflare has launched Cloudflare OS, an open-source browser workspace that lets employees build and share AI-powered tools using their company’s own data and systems.

According to reports, the platform is designed to move AI work beyond centralized developer teams, allowing staff to research, create live-data documents, and automate routine workflows without coding. Companies run it in their own Cloudflare accounts, keeping control of the apps, connections, and workplace context they create.

NotebookLM is dead.

(We’ll miss you, dear friend.)

What replaced it might be the easiest AI win your company grabs all year.

Google graduated its cult-favorite research tool into Gemini Notebook, and has rolled out seven upgrades that turn it agentic by default.

Translation: it stopped waiting for orders and started doing real work. The goodies locked behind Google's priciest plan just hit every paid account, prolly yours too.

Don't worry: none of it needs code or engineers.

That's what we tackle on today's episode of Everyday AI: what the rebrand changed, the one-prompt move that builds five deliverables mid-meeting, and the grounding shift that could burn careless teams.

It's Wednesday, y'all, so let's put AI to work.

1. NotebookLM dies, Gemini Notebook goes full agent 🚀

Everybody assumed this tool was next for the Google graveyard, but instead it got promoted into the flagship Gemini family. That promotion matters way more than the new logo: you can finally build serious workflows without fearing a random shutdown email.

Under the hood, the changes run a whole lot deeper than the name suggests. Old NotebookLM ran a non-reasoning model with a fixed output menu, while the new one runs Gemini 3.5 reasoning, Google's Antigravity harness, and a secure cloud computer in every notebook.

So why should busy leaders care? Because your notebook writes and executes code for research, calculations, and data analysis, grounded in YOUR sources, not the open internet.

Try This

Drop something real into a fresh notebook this week, meeting notes or a market report, and ask one question that needs calculation, not just a summary.

Then open the thoughts panel and watch it reason, run code, and cite your sources in real time, a free masterclass in how agents think.

Just 10 minutes there beats a month of headlines.

2. One prompt now builds five executive deliverables ⚡

This is where agentic stops being a buzzword and starts printing time back into your week. We threw it a beefy live test: analyze three AI budget strategies for a 5,000 person enterprise, run the cost math, pick a winner, and then build five deliverables in one run.

What came back? A custom audio overview, a mind map, an editable Excel workbook with working formulas, a Nano Banana infographic, and a two-page executive brief, every claim traceable to the source notes.

The whole package landed in about six minutes, versus an afternoon of assembly across five tools, and that kind of math changes how fast your whole team ships decisions.

Try This

Grab one real decision off your desk, feed the background docs into a notebook, and ask for the whole package in one prompt: analysis, spreadsheet, exec brief, and infographic.

Then open the Excel workbook it hands back, because yes fam, the formulas actually work.

Once your most skeptical teammate watches a finished package appear mid-meeting, the buy-in conversation ends itself.

3. The 100% grounding promise just got an asterisk 🔥

Quick context: executives trusted NotebookLM because it answered from your sources ONLY, crushing hallucinations in a way OpenAI, Anthropic, and Microsoft never matched, and Gemini Notebook keeps that grounding by default.

Now here's the gnarly part, though. When your sources come up short, it can offer to hop online and pull outside info into your notebook, and once you say yes, outside data starts mixing with yours.

The guardrail is real here, at least for now, since it always asks your permission before searching anything.

But if your team assumes every answer still comes purely from your documents, that assumption became conditional, and leaders gotta know which mode they're in.

Try This

Stress test this yourself by asking a notebook something your sources definitely can't answer, then watch it flag the gap and request permission first before touching the web.

After that, set one line of team governance that keeps the hallucination protection you signed up for: grounded by default, web only with approval, chain of thought checked when stakes are high.

 

Reply

or to participate.