Google made two announcements Thursday that, taken together, mark the clearest signal yet that the company is transforming its products from information surfaces into action layers: it renamed its AI research tool NotebookLM to Gemini Notebook, and it opened AI Mode in Search to bidirectional integration with Instacart, Canva, and YouTube Music — meaning Search can now not just surface information but write actions directly into apps you use every day. If you use Google Search and have a connected Instacart account, what changed today affects how you shop, plan, and create.
NotebookLM Is Now Gemini Notebook — What Actually Changed
NotebookLM, which debuted at Google I/O 2023 under the codename Project Tailwind, became Gemini Notebook on Thursday. According to Google’s official Gemini Notebook announcement, the new name reflects the product’s integration into the Gemini family, though Google says it will remain a standalone research tool rather than merging into the general-purpose Gemini assistant. According to Josh Woodward, VP of Google Labs, the product now serves more than 30 million users and 600,000 organizations — figures Google reported without independent verification.
The rebrand is not the substantive story. That story happened on June 8, when Google overhauled the product’s underlying architecture: it upgraded from Gemini 1.5 to Gemini 3.5, integrated Antigravity (Google’s agentic orchestration infrastructure, introduced at I/O 2026), and added a dedicated secure cloud computer to every notebook. XDA Developers’ hands-on with the update offers a detailed look at what changed. What that means for users is explained in the next section.
The new name and updated Gemini gradient logo are rolling out across the product’s interfaces over the next several weeks. Mobile users may need to update their apps. Existing shared notebook links and URLs will redirect automatically — Google says no action is required from admins or end users.
How the Cloud Computer Inside Every Notebook Actually Works
The most technically significant part of today’s announcement is the expansion of the June 8 cloud-computer capability to AI Pro subscribers, who had been excluded from the initial Ultra-only rollout. Understanding what that capability actually does requires distinguishing it from what NotebookLM previously did.
The prior version of NotebookLM was a retrieval-augmented generation system: it indexed the documents you uploaded into a vector store, retrieved relevant passages in response to your queries, and synthesized those passages into answers. It could tell you what a spreadsheet said; it could not compute on the spreadsheet. The new Gemini Notebook does something fundamentally different.
Each notebook now has access to a sandboxed cloud computer — an isolated execution environment running on Google Cloud’s code execution platform. The sandbox creates a secure, isolated space for running code, and it supports a library of more than 100 curated software skills. When you ask Gemini Notebook to analyze a dataset, it writes Python code, executes it inside that isolated container, and returns structured results — charts, tables, or full output documents — rather than just summarizing the data in prose.
The Antigravity orchestration layer is what connects the reasoning model to the execution environment. Gemini 3.5 Flash plans the analysis, writes the code, and interprets the results; Antigravity manages the multi-step task sequence, handles state across turns, and routes outputs back to you. The result is a tool that can produce PDF reports with charts, Excel spreadsheets, and PowerPoint decks without leaving the notebook interface. Google’s internal benchmarks claim the upgraded system outperformed its predecessor in more than 65 percent of side-by-side evaluations, with the margin reaching 78.2 percent on advanced web research tasks and 69.9 percent on large-document analysis. These are Google’s own figures, derived from internal evaluations — no independent auditor has verified them, and readers should treat them as directional rather than definitive.
Dan Fitzpatrick’s Forbes analysis identified a real risk that comes with this power: when Gemini Notebook selects the analysis method, writes the code, and explains the result, users can accept AI-generated outputs without examining the underlying methodology or assumptions. That is a meaningful caveat for anyone using the tool for research, data analysis, or student coursework — the output looks authoritative even when the method is wrong.
The cloud-compute capability is live now for Google AI Ultra subscribers and Workspace customers with AI Ultra Access or AI Expanded Access. It will reach all AI Pro users on the web over the coming weeks.
How Does Google Search Know What to Add to My Instacart Cart?
AI Mode in Search — the conversational search interface that launched in early 2025 and has been expanding throughout 2026 — gained the ability Thursday to execute tasks inside Instacart, Canva, and YouTube Music, according to Google’s connected apps announcement. The integration is US-only at launch and works on both desktop and mobile.
The architecture builds on Google’s “Personal Intelligence” and “Connected Apps” framework, which has been part of the Gemini app since Google I/O 2026 and arrived in AI Mode in Search earlier this year. TechCrunch’s AI Mode integration report details how the feature expands beyond the Gemini app. The prior system was read-only: AI Mode could pull context from your Gmail, Calendar, and connected apps to personalize responses. What changed Thursday is that the data flow is now bidirectional — Search can write actions back into third-party apps, not just read context from them.
Here is how each integration works in practice. If you are planning a barbecue and build a grocery list through an AI Mode conversation, connecting your Instacart account allows those ingredients to be added directly to your Instacart shopping cart; you then open the Instacart app or website to complete the checkout. For Canva, if you describe a design project — creating party flyers, for example — AI Mode can surface relevant Canva templates, drawing on connected calendar context to inform what it pulls up. For YouTube Music, a playlist assembled through an AI Mode conversation can be saved directly to your account and played immediately.
The Instacart integration is the most meaningful for privacy, because it involves writing structured commercial data (items and quantities) derived from a conversational AI session into a shopping cart that may be tied to your purchase history, saved payment methods, and preferred store locations. Google says users must explicitly connect each app in AI Mode’s Connected Apps settings, and that data is handled under the Gemini privacy policy. That policy notes that some data from Gemini sessions may be accessed by human reviewers for safety and quality purposes — a condition that applies regardless of which apps are connected.
What Happens When Search Takes Action on Your Behalf
The conceptual shift here is worth naming clearly, because the product announcements from Google frame it as a convenience upgrade rather than a structural change.
Today’s AI Mode app integrations turn Google Search into a transactional AI agent for a subset of use cases. Prior to Thursday, AI Mode was a sophisticated information retrieval and synthesis tool. After Thursday, it is a system that can initiate commercial and creative actions on your behalf — adding items to a shopping cart, generating design templates, curating playlists — based on conversations that involve your calendar data, your stated preferences, and your account credentials at third-party services.
The question of who is accountable when a transactional agent acts incorrectly is not addressed in Google’s launch materials. If AI Mode adds the wrong items to your Instacart cart and you do not review the cart before checking out, the error is yours to remediate with Instacart. The Gemini Apps Privacy Hub notes that the system may share with connected services “data like your name and address or info you find sensitive.” The scope of what counts as “necessary information” for a grocery-list action is defined by Google’s own implementation, not by a published technical specification accessible to users.
Google’s approach requires connecting apps through a deliberate opt-in step, which provides some protection against accidental scope creep. But the consent model is account-level (you connect Instacart once) rather than per-transaction — meaning once the connection exists, every AI Mode conversation can write to your cart without a separate authorization step for each action.
For the far larger question of what Google learns about your preferences, shopping behavior, and plan-making from these interactions, the answer sits inside the Gemini privacy policy rather than in a clear disclosure at the point of connection.
What This Means for the Competitive Landscape
Google’s dual announcement reflects a competitive dynamic that has accelerated sharply in 2026. OpenAI’s ChatGPT and Anthropic’s Claude both support third-party app integrations, and the race to build AI systems capable of completing real-world tasks — not just answering questions — has pushed every major AI provider toward deeper ecosystem hooks, as TechCrunch’s competitive landscape analysis notes.
Google’s approach threads a specific needle. Rather than routing all user intent through a single general-purpose assistant, the company is converting its highest-traffic existing surfaces — Search, the Gemini app, Workspace — into action layers, while positioning Gemini Notebook as a specialized deep-research tool for users who need it.
A product confusion problem remains, and several reviewers flagged it Thursday: Google’s Gemini app also has “Notebooks” — a feature for organizing long-running projects with persistent context. Those notebooks sync with Gemini Notebook, but they are not the same product. One is a chat-organization feature inside a general assistant; the other is a standalone document-analysis and code-execution environment. The products share a name and data, but serve different use cases. Google has not published clear guidance distinguishing them beyond positioning Gemini Notebook as the “premier research tool.”
For the 30 million people already using Gemini Notebook, today’s announcement delivers a meaningful expansion of what their existing tool can do — the code-execution capability rolling out to AI Pro subscribers this month adds genuine analytical power that the prior NotebookLM could not touch. For the much larger population of daily Google Search users, the third-party app integrations represent a decision point: connecting your Instacart, Canva, or YouTube Music account to AI Mode adds convenience, but it also extends Google’s data reach into your commercial and creative workflows in ways that deserve deliberate consideration before connecting.
Frequently Asked QuestionsWhat is Gemini Notebook and how is it different from the old NotebookLM?
Gemini Notebook is the new name for NotebookLM, effective July 16, 2026. The product has a new name and logo but the same core function: a standalone AI research tool that lets you upload documents and query them with an AI assistant. The substantive change happened on June 8, when Google upgraded the underlying architecture from Gemini 1.5 to Gemini 3.5 and added a sandboxed cloud computer to every notebook. Where the old NotebookLM could only read and summarize your documents, the new Gemini Notebook can write and execute Python code against them, produce structured outputs like Excel files and PDF reports with charts, and perform complex data analysis — not just explain it in prose.
Do I need a paid plan to use Gemini Notebook’s code execution features?
Yes. The cloud-computer and code-execution capability is currently available to Google AI Ultra subscribers and Workspace customers with AI Ultra Access or AI Expanded Access. Google says it will expand access to all AI Pro subscribers ($20/month) on the web over the coming weeks. The standard free tier does not include these features; the timeline for free-tier access, if any, has not been announced.
What data does Google access when AI Mode writes to my Instacart or Canva account?
When you connect Instacart, Canva, or YouTube Music to AI Mode, you authorize Google’s Connected Apps framework to pass structured data to those services on your behalf. For Instacart, that includes the item names and quantities generated in your AI Mode conversation. The Gemini privacy policy states that to complete tasks through Connected Apps, Gemini may share with third-party services “necessary info, which can include data like your name and address or info you find sensitive” — and that “some data may be accessed by human reviewers” for safety and quality purposes. You can manage connected apps in AI Mode’s settings and disconnect them at any time, but the account-level connection means the system can write to your cart in any AI Mode session without a per-transaction authorization step. Google does not publish a technical specification of exactly what data fields are shared in each integration.
Is the AI Mode Search integration the same as what was already in the Gemini app?
Largely yes, architecturally. Google introduced Canva, Instacart, and OpenTable integrations in the Gemini app earlier in 2026 through the same Connected Apps framework. Thursday’s announcement brings that same capability to AI Mode in Google Search — the search interface itself — which reaches a much larger daily-active user base than the dedicated Gemini app. The Search integration is currently US-only; the Gemini app integrations launched earlier and cover a broader set of partner apps.