NotebookLM’s new agentic AI update turns it into a real research assistant

NotebookLM just made a huge leap. What started as a simple “chat with your documents” tool is quickly turning into a full agentic research assistant that can discover sources, analyze data, execute code, and even generate finished reports and slide decks for you.

From chat with PDFs to agentic research workspace

Previously, NotebookLM was most useful once you had everything ready: PDFs, links, notes, and other sources. You uploaded your files, asked questions, and got summaries or explanations back.

With the new update, NotebookLM is evolving into a true research workspace. It can now help you:

• Discover and add relevant sources from the web
• Reason across multiple documents and data sources
• Run multi-step research workflows automatically
• Execute code and perform deeper analysis in a secure cloud environment
• Turn your research into structured, downloadable outputs

In other words, it’s starting to behave more like an AI agent that actually does work for you, not just a chatbot that answers questions. This push toward agents lines up with broader trends across the AI ecosystem, similar to what we’ve seen in updates like agent-focused tools that make AI actually useful for real workflows.

Powered by Gemini 3.5 Flash and Antigravity

The new NotebookLM experience is now powered by Gemini 3.5 Flash and Google’s Antigravity stack. That means faster responses, better reasoning, and more visibility into how the model is thinking through tasks.

Google shared early benchmark numbers showing that the new system beats the previous NotebookLM setup over 65% of the time across key areas:

• Accuracy and quality
• Multilingual support
• Large document analysis
• Document creation
• Advanced research

For large document analysis, the new system had a 69.9% win rate. For advanced web research and source directory tasks, it reached a 78.2% win rate versus the old baseline. So this isn’t just a minor quality-of-life update—it’s a substantial performance jump.

Each notebook now has its own secure cloud computer

One of the biggest changes is that every notebook now comes with a secure cloud computer attached to it. This environment includes over 100 curated software skills and tools that NotebookLM can call on during a task.

Instead of only summarizing PDFs or answering simple questions, NotebookLM can now:

• Run code for deeper analysis
• Handle multi-step workflows (like “clean this data, analyze it, then visualize the results”)
• Use specialized tools inside the notebook to complete complex research tasks

This is where NotebookLM starts to feel like a lightweight coder and research agent. You can give it messy data, complex documentation, or vague research questions, and it can use its toolset to work through the problem end-to-end.

Agentic research: start from ideas, not just files

Previously, you needed a well-organized notebook full of sources before NotebookLM became really useful. Now, you can start with almost nothing—just a loose idea, a question, or a rough research direction.

NotebookLM can:

• Suggest relevant topics and angles to explore
• Find high-quality sources from the web using Google Search
• Propose which sources to add to your notebook (with your permission)
• Help you build a research base from scratch

Crucially, you stay in control. NotebookLM won’t just dump sources into your notebook. It will surface candidates, ask if you want to add them, and then attribute them properly once they’re included.

This makes the early research phase far less intimidating. If you’re exploring a new topic, trying to see different perspectives, or hunting for primary sources in another language, NotebookLM can now help you shape the entire research direction—not just summarize what you already have.

Structured outputs: from notes to finished deliverables

Another major upgrade is NotebookLM’s ability to turn your sources into real, downloadable outputs. Instead of just giving you text answers in chat, it can now assemble context from across your sources and generate full files.

Supported output formats include:

• PDFs and Word documents
• Markdown reports
• Excel sheets and CSVs
• PowerPoint slide decks
• JSON files
• Images generated via Nano Banana (for charts or visuals)

This effectively turns NotebookLM into a tool that can help you go from raw information to final deliverables. Some example workflows:

• Give it business documents and ask for a polished slide deck
• Upload raw data and ask for charts, spreadsheets, and clean structured tables
• Combine multiple research papers into a single, well-organized report

Instead of copying and pasting answers out of chat, you can now ask NotebookLM to produce the actual files you need.

Source attribution for every artifact

Trust is a big issue with AI-generated content, and Google is addressing that directly. When NotebookLM generates a report, chart, slide deck, or any other artifact, you can see:

• Which prompts were used
• Which sources were referenced
• How the output ties back to your materials

This source attribution makes it much easier to verify claims, follow up on references, and understand how the AI arrived at its conclusions. You’re not just handed a mysterious AI file—you can trace it back to the underlying data.

How researchers can use the new NotebookLM

For researchers and students, the update unlocks a lot of new workflows:

• Bring in messy data from multiple sources and let NotebookLM clean and normalize it
• Run analysis across different datasets or papers and get charts plus a written summary
• Discover related work, primary sources, and cross-language materials
• Generate a final report or literature review as a downloadable document

Because NotebookLM can now both discover sources and analyze them, it’s much closer to a full research partner than a simple summarization tool. It also fits neatly into the broader wave of AI tools aimed at advanced research and analysis, similar to what’s covered in updates like recent Gemini and NotebookLM upgrades.

Using NotebookLM as a lightweight coding and technical assistant

The new secure cloud computer and code execution features make NotebookLM surprisingly useful for technical teams as well.

Some practical uses:

• Feed it complex technical documentation (APIs, libraries, frameworks) and have it produce a concise markdown guide you can download
• Ask it to explore new language features or packages, then summarize the key points
• Turn dense engineering specs into simplified guides, implementation plans, or slide decks
• Use it as a bridge: generate a clean, structured spec or dataset in NotebookLM, then hand that off to your coding agent or IDE assistant

While it’s not a full-blown coding environment like a dedicated AI coding tool, the ability to execute code as part of a research workflow makes it feel like a lightweight coder that can help with analysis, prototyping, and documentation.

Small business and operations use cases

For small businesses and operations teams, NotebookLM can now act as a flexible analytics and reporting assistant.

You can, for example:

• Upload campaign data, sales numbers, and ad spend
• Ask NotebookLM to calculate what worked, what didn’t, and whether a campaign is worth scaling
• Have it generate charts, tables, and a written summary of performance
• Export everything as a report, spreadsheet, or slide deck for your team

This turns scattered business data into clear insights and ready-to-share deliverables, without needing a dedicated analyst or complex BI stack.

Future: better video generation from your notebooks

There’s also a hint of what’s coming next. NotebookLM’s current video capabilities are limited, but Google is expected to integrate a new Omni video generation model from Gemini directly into the product.

Once that lands, you could potentially:

• Turn your research notebook into infographic-style videos
• Generate simple animations that explain your findings
• Create visual explainers based on your structured outputs and charts

That would push NotebookLM even further—from research and reports into full multimedia content creation.

Availability and what this update means

The new agentic research features are rolling out first to Google AI Ultra subscribers, with broader availability planned for all paid plans soon.

Overall, this update transforms NotebookLM from a passive Q&A tool into an active, agentic collaborator. It can help you:

• Start from vague ideas and build a full research base
• Discover, clean, and analyze data across sources
• Execute code and use specialized tools in a secure cloud computer
• Produce final, trustworthy artifacts with clear source attribution

If you’ve previously thought of NotebookLM as just a place to dump PDFs and ask for summaries, it’s time to revisit it—because it’s quickly becoming one of the most capable AI research workspaces in Google’s ecosystem.