Google said on June 8, 2026 that it is rolling out a major NotebookLM upgrade, moving the product onto Gemini 3.5, adding agentic source discovery in chat, giving each notebook a secure cloud computer for code execution, and expanding output formats to include charts, PDFs, spreadsheets, structured data, and PowerPoint files. The update is available globally on the web starting today for Google AI Ultra users and Workspace business customers with AI Ultra Access or AI Expanded Access.
What changed in NotebookLM on June 8
The biggest product shift is that NotebookLM no longer depends as heavily on users arriving with a finished source set. Google says users can now start with loose ideas and questions, then use chat to build a source repository directly inside the notebook. The system can use Google Search to suggest relevant web sources, while keeping the user in control of which materials get added.
Google also moved NotebookLM to Gemini 3.5 and Antigravity, which it says improves reasoning quality and gives users better visibility into how the system is working through research tasks. Each notebook now includes a secure cloud computer that can write and run code, and Google says the system comes with more than 100 curated software skills for deeper analysis.
On the output side, NotebookLM now supports a wider set of downloadable artifacts. Google says users can generate data visualizations and charts, PDF and Word-style documents, CSV and JSON files, Excel workbooks, PowerPoint decks, and image outputs. TechCrunch reported that Google is also surfacing more detailed intermediate steps in chat so users can inspect how NotebookLM reached an answer.
Why this matters more than a feature refresh
NotebookLM started as a grounded research assistant: useful, but still dependent on manual source gathering and relatively narrow output patterns. This update pushes it closer to an agent-style knowledge workflow. The product can now help find inputs, analyze them with code, and package results into business-ready deliverables without forcing users to jump across multiple tools.
That matters because the competitive line in AI is moving away from raw model access and toward end-to-end workflow completion. A research assistant that can discover sources, reason over long documents, run analysis, and export a slide deck or spreadsheet is much closer to how real analysts, operators, and managers work than a simple chatbot summary experience.
Google’s own evaluation numbers reinforce that point. The company says the upgraded NotebookLM achieved an average win rate of more than 65% against its prior system across five core evaluation dimensions, including a 69.9% win rate in large document analysis and a 78.2% win rate in advanced web research and source discovery. Even if those are internal benchmarks, they show where Google believes the product is improving fastest.
Business impact for AI agents and knowledge teams
The clearest near-term winners are teams that spend large amounts of time collecting, checking, and reformatting information before they can act on it. Research, strategy, product, operations, and customer-facing knowledge teams often lose time not on final writing, but on building a trustworthy source base and converting it into usable outputs. NotebookLM is now aimed much more directly at that bottleneck.
For enterprises, the important signal is not just that NotebookLM can answer questions better. It is that Google is bundling grounded retrieval, code execution, structured output generation, and guided source discovery into one controlled environment. That makes NotebookLM feel less like a note-taking companion and more like a research execution layer.
For the broader AI agent market, the June 8 release is another reminder that agent competition is increasingly about packaging. The model still matters, but the differentiator is how well a product can turn messy context into reliable work products with inspection, attribution, and editing built in. In that sense, NotebookLM is moving closer to the kind of role-specific AI worker businesses actually buy.
What to watch next
The next question is whether Google widens access quickly and whether NotebookLM’s new workflow features stay reliable outside controlled demos and internal evaluations. If the source-discovery flow remains grounded and the code-running environment proves useful on messy real-world documents, NotebookLM could become a stronger business research surface inside the broader Google AI stack.
The practical takeaway for AI agents and automation teams is straightforward: research workflows are becoming more agentic, not just more conversational. Businesses evaluating internal knowledge agents should now expect tools to help assemble source context, run analysis, and produce usable deliverables in one loop rather than stopping at summarization.