In a massive expansion of its beloved AI note-taking tool, Google has rolled out a sweeping set of upgrades that evolve NotebookLM from a passive, document-grounded reading assistant into an autonomous, full-scale Research Agent.

Historically, NotebookLM operated as a strict “closed-loop” system—you brought your own documents, and it answered questions based solely on that text. The 2026 upgrade shatters that limitation by integrating live web research, advanced agentic coding capabilities, and an array of downloadable, corporate-ready export formats.

1. Upgraded Foundations: Gemini 3.5 and Visible Reasoning

The backend engine powering NotebookLM has been upgraded to Gemini 3.5 and Antigravity, Google’s specialized multi-agent coding framework.

• Thinking Transparency: Responding to user feedback demanding clearer insights, NotebookLM’s revamped chat interface now explicitly lays out its step-by-step reasoning and expanded thinking process directly on screen as it works through a prompt.
• Massive Performance Leaps: In side-by-side internal testing against its previous architecture, Google reports that the new agentic model achieved a 69.9% win rate in large document analysis and an exceptional 78.2% win rate in advanced web research and source discovery.

2. Flipping the Dynamic: Start with an Idea, Not a File

The biggest philosophical shift in the new NotebookLM is how projects get off the ground.

[Old Workflow] You must upload PDFs/Docs first ──► AI analyzes your data
                                                               
[New Agentic Workflow] Start with a loose idea/question ──► AI searches web ──► Proposes curated sources ──► You approve repository

Instead of requiring a ready-made library of personal documents to get started, you can now enter a workspace with nothing but a loose idea, a question, or a research goal. NotebookLM acts as an autonomous scout, leveraging Google Search to scrape the web, discover relevant background information, and propose primary sources.

The Privacy Guardrail: True to its original promise, the user remains the ultimate editor. NotebookLM suggests web discoveries and attributes them transparently, but nothing is formally added to your permanent notebook repository without your explicit review and approval.

3. The Code Engine: A Virtual Machine in Every Notebook

In an upgrade that positions the platform to take on heavy data-science workflows, every single notebook workspace is now equipped with a secure cloud computer (virtual machine).

By leaning on the Antigravity integration, NotebookLM can autonomously write and execute python code on your behalf to process information. Backed by a library of over 100 pre-packaged software skills, the agent can split complex analytical projects into smaller tasks, spinning up parallel AI agents to clean messy datasets, standardize spreadsheet layouts across disparate sources, or parse mathematical metrics directly from your uploaded content.

4. The Studio Panel: High-Fidelity Professional Exports

NotebookLM has evolved beyond text summaries and the viral “Audio Overview” podcasts. A newly expanded Studio Panel functions as an export factory, transforming raw source research into fully formatted, boardroom-ready downloads:

• Data Visualizations: Generates native charts and graphs downloadable as PNG or SVG files.
• Office Suites: Exports directly into standard professional formats including Microsoft PowerPoint (.pptx), Microsoft Excel (.xlsx), Word (.docx), and PDFs.
• Structured & Creative Outputs: Spits out clean JSON/CSV code blocks for external applications, alongside student worksheets, project roadmaps, and infographics.
• Multilingual Flow: The agent fully supports cross-lingual research. You can instruct it in English to scan complex, primary sources written in Mandarin or German, and it will output the finalized Excel budget or PDF report entirely in English.

Access and Availability

The agentic upgrades are rolling out globally on the web. To manage the immense computational overhead required to run virtual code environments and deep web research, Google is staggering access based on subscription tiers:

• Eligible Users: Access is available immediately for consumers on Google’s AI Ultra plan (the $100/month tier introduced at I/O 2026), alongside enterprise teams utilizing Workspace AI Ultra Access and AI Expanded Access add-ons.
• Data Security: Google re-emphasized that for all Workspace and enterprise accounts, uploaded data, custom queries, and agentic outputs are completely bound by enterprise-grade protection—meaning your research remains locked within your project and is never reviewed by humans or used to train foundational AI models.

Note: While the web overhaul is live, the new web-search and code-execution tools are not yet supported on the NotebookLM mobile app.