From AI Note‑Taking to Agentic Research Partner
NotebookLM Gemini 3.5 is Google’s upgraded, source‑grounded AI research environment that combines advanced reasoning, agentic research tools, and cloud-based AI code execution to transform traditional AI note-taking into end-to-end research automation across documents, web sources, and data workflows. Originally designed to help users organize and explore uploaded materials, NotebookLM now starts from open-ended questions, finds sources with Google Search, and builds a curated notebook on the fly. The tool still anchors answers in user‑approved sources, keeping citations visible so you can check claims against primary materials. What changes is the level of initiative: instead of waiting for you to feed it PDFs and links, NotebookLM proposes relevant research directions, suggests overlooked papers or primary sources, and surfaces connections across your library. The result is a shift from a passive assistant that summarizes notes into an active research agent that plans, searches, analyzes, and delivers finished outputs.
Gemini 3.5 and Antigravity Bring Visible Reasoning
At the core of this upgrade is the move to Gemini 3.5 and Google’s Antigravity tool layer, which power NotebookLM’s new reasoning depth. The chat experience runs on the latest model, with expanded, step-by-step thinking shown directly in the interface so you can trace how the system reached its conclusions. According to Google’s own evaluations, “the upgraded NotebookLM achieved an average win rate of over 65% across our top five core evaluation dimensions,” including a 69.9% win rate in large document analysis and 78.2% in advanced web research and source discovery against the prior baseline. For complex research projects, that means better handling of long reports, multi-document comparisons, and nuanced queries. Antigravity adds the ability to call specialized tools in the background, so when you ask a dense question, NotebookLM can search, inspect sources, run analyses, and then explain its reasoning path back to you.
Secure Cloud Code Execution Turns Notebooks Into Labs
Every notebook now includes a secure cloud computer, turning NotebookLM into a lightweight research lab where AI code execution is part of the workflow. Instead of exporting data to external tools, you can ask NotebookLM to write and run code to analyze datasets, test hypotheses, or generate visualizations, all grounded in the sources in your notebook. Google says the system provides more than 100 curated software skills, allowing the agent to select from tools for data analysis, plotting, text processing, and more. A single question can trigger source discovery, data extraction, a calculation, and a chart without leaving the notebook. This matters for researchers who previously had to juggle a notebook app, IDE, and office suite: NotebookLM now threads those steps together, reducing context switching and making it easier to audit both the code and the evidence behind any generated result.
From Answers to Artifacts: Reports, Charts, and Slides on Demand
NotebookLM’s export upgrades push it further into full research automation. Once analysis is done, you can ask it to generate finished outputs instead of raw text. The system can now create reports with charts and tables, data visualizations (png, svg), structured data (csv, json), and office-ready files like PDFs, docx, markdown, xlsx, and pptx. You can give detailed export instructions, review what NotebookLM produces, and iterate without rebuilding from scratch. That workflow turns a conversational session into polished research artifacts: literature reviews grounded in notebook sources, budget summaries with linked spreadsheets, or presentation decks assembled from your findings. Because sources stay attributed, each artifact remains auditable; you can trace each chart or claim back to specific documents or web pages. The combination of reasoning, AI code execution, and export options means NotebookLM now spans the full path from question to shareable output.
Who Gets Access First—and How Expectations Will Shift
Google is rolling out these capabilities on the web first to Google AI Ultra subscribers and Workspace customers with AI Expanded access, with broader availability planned later. Early adopters are likely to be schools, universities, researchers, and workplace teams already embedded in Google’s productivity stack. For them, NotebookLM stops being a sidecar note app and becomes a central research agent that can initiate work, not only respond. That shift will raise expectations for AI note-taking tools in general: users will look for systems that can find sources, explain reasoning, execute code, and deliver ready-to-share outputs, all while keeping them in control of the source library. It also puts more pressure on transparency and evaluation, since Google’s impressive win-rate figures are company claims rather than independent validation. As access widens, real-world use will decide whether this new agentic model becomes the research default.
