Google expands NotebookLM with “agentic” research upgrades, code execution, and new document formats

Google on Monday rolled out across-the-board upgrades to NotebookLM, positioning the product as a more capable “research partner” for complex projects. The updates, announced in a Google Blog post dated June 8, 2026, add new agent-like capabilities in chat, more advanced reasoning, and new ways to turn source material into downloadable research artifacts.

NotebookLM was launched about three years ago as an experimental product from Google Labs, according to Google. The company said millions of people and organizations use it as a collaborative knowledge and research tool, helping users organize their thinking, find deeper connections across documents, and generate new ideas based on the sources they select.

Google said it is upgrading NotebookLM’s underlying chat experience to run on Gemini 3.5 and an additional component it calls “Antigravity,” describing the change as improving accuracy and reliability and providing better visibility into how the system is reasoning. Each notebook also now includes what Google describes as a secure cloud computer, which is intended to let NotebookLM write and run code for deeper research and more complex analysis. Google added that the setup includes more than 100 “curated software skills,” designed to unlock additional capabilities for working with and understanding sources.

In internal side-by-side evaluations against a prior system, Google reported performance gains for the upgraded NotebookLM. It said the new system achieved an average win rate of over 65% across its top five evaluation dimensions, with a 15 percentage-point margin above parity. The post also cited specific improvements, including a 69.9% win rate for large document analysis and a 78.2% win rate for advanced web research and source discovery.

A major part of the update is NotebookLM’s expanded ability to produce structured outputs. Google said users can now ask NotebookLM to create outputs in more formats, with the system assembling context from the user’s sources to generate deliverables such as PDF reports with charts and tables, detailed budget spreadsheets, and bespoke student worksheets. The blog post listed new output formats, including data visualizations and charts (PNG and SVG), documents (PDFs, DOCX, Markdown, and text files), images generated under a feature Google refers to as “Nano Banana” (PNG, JPG, and GIF), structured data (CSV and JSON), and Microsoft file types including Excel (XLSX) and PowerPoint (PPTX). Google also said users can edit outputs after they are generated, and that additional formats are planned for the future.

Google also said the product is becoming easier to start with when users do not yet have a well-formed question or a ready-made set of sources. Previously, Google said NotebookLM was most useful when people came with their own materials. In the new approach, NotebookLM can guide users through building a “source repository” within the chat, and it can use Google Search to find relevant, high-quality sources from the web to add to the notebook. Google emphasized that users remain in control of which sources are added, and that sources continue to be clearly attributed.

The post does not disclose several details that would matter to enterprise buyers and power users, including which specific “software skills” are included in the secure cloud computer environment, what limitations apply to code execution (such as timeouts, supported languages, or compute caps), or how broadly the new artifact-generation feature is supported across different NotebookLM tiers. Google also did not provide pricing changes or explain whether availability differs by industry or region, beyond the high-level access categories it listed.

Looking ahead, the main question is how quickly the updates spread beyond the initially covered users and whether the new workflows change how organizations use AI for research, reporting, and analysis. Google said the upgrades are rolling out globally on the web starting June 8, 2026 to users with Google AI Ultra and certain Google Workspace business customers with AI Ultra Access and AI Expanded Access, with plans to expand to others over time. Observers will likely watch for additional output formats, further expansion of the rollout, and feedback on the quality of source-grounded outputs as Google iterates.