NotebookLM 2.0 Agentic Update: Everything That Changed in June 2026

The Biggest NotebookLM Update Ever

On June 8, 2026, Google shipped what is easily the largest NotebookLM release since the product launched. Internally referred to as the "2.0 agentic update," it turns NotebookLM from a smart research reader into an active research agent — one that can run code, search the open web, discover its own sources, and export polished, editable files in nearly every format an office worker touches.

If you've only ever thought of NotebookLM as "the thing that makes AI podcasts," this update is your cue to look again. Below is a complete breakdown of everything that changed, why it matters, and — since this is a watermark-removal site — what all these shiny new export formats mean for anyone who wants clean, professional files.

1. Gemini 3.5 and the Antigravity Framework

At the core of the update is a new model, Gemini 3.5, running on Google's Antigravity agent framework. Antigravity is the orchestration layer that lets the model plan multi-step tasks, call tools, execute code, and check its own work rather than answering in a single pass.

In practice, this is the difference between asking a librarian for a book and asking a research assistant to go read six books, run the numbers, and come back with a written report. NotebookLM no longer just retrieves and summarizes — it reasons through a task, breaks it into steps, and uses tools to complete each one. Every capability below is built on top of this foundation.

2. Every Notebook Gets a Secure Cloud Computer

This is the headline feature. Every notebook now comes with its own sandboxed cloud computer capable of running code. Ask NotebookLM to analyze the numbers in a source document, and instead of guessing at the math the way language models used to, it writes and executes actual Python, then returns verified results.

What this unlocks:

• Real data analysis — parse a spreadsheet, compute statistics, run correlations, and generate charts from your actual data.
• Accurate calculations — no more hallucinated arithmetic; the computer does the math.
• Data transformation — clean, reshape, filter, and pivot datasets before turning them into outputs.
• Visualizations — generate plots and charts directly from source data.

The sandbox is isolated and ephemeral, so it's secure by design — code runs in a throwaway environment, not on your machine or against your Google account.

3. Start-From-Scratch Research Mode

Until this update, NotebookLM could only work with sources you uploaded. That was its defining constraint — and its defining strength (no hallucinations from outside the source set). The 2.0 update adds an optional start-from-scratch research mode that flips this on its head.

Give NotebookLM a topic instead of a document, and it will:

• Search the open web for relevant, high-quality material.
• Discover and propose sources — articles, papers, and pages you didn't know existed.
• Assemble a source set for you, which you can then curate before it generates anything.

Crucially, you stay in control: the model proposes sources, you approve them. This keeps the grounding benefit (everything traces back to a real, visible source) while removing the "empty notebook" cold-start problem. It's the single biggest change to how you begin a research project in NotebookLM.

4. Massively Expanded Export Formats

Here's where the update gets directly relevant to this site. NotebookLM used to export a handful of formats — mostly PDF slides, audio, and video. The 2.0 update dramatically widens the export menu, and much of it is now fully editable:

• Editable PPTX — real PowerPoint files with movable text boxes and shapes, not flattened PDFs.
• XLSX spreadsheets — structured data output straight into Excel/Sheets.
• DOCX documents — editable Word documents for reports and write-ups.
• CSV files — raw structured data for import into any tool.
• Charts and visualizations — generated from the code-execution sandbox described above.

The move from "PDF only" to editable PPTX in particular closes one of NotebookLM's most-complained-about gaps. We wrote about that exact frustration in our NotebookLM limitations guide — and this update fixes it at the source.

5. 100+ Curated Software Skills

Riding on the code-execution sandbox, Google added a library of over 100 curated "skills" — pre-built capabilities the agent can invoke for common tasks. Think document formatting, data cleaning, chart styling, statistical routines, and file conversions, all packaged so the model reliably reaches for the right tool instead of improvising.

For users, this means outputs are more consistent and professional-looking. You don't need to know that a task requires a particular library or technique; the skill handles it, and NotebookLM picks the right one for the job.

6. Transparent Chain-of-Thought Reasoning

With agents doing multi-step work, trust becomes a real concern — you need to see how a conclusion was reached. The update surfaces transparent chain-of-thought reasoning: you can watch NotebookLM plan its approach, see which sources it consulted, and follow the steps it took to reach an answer.

This isn't just a nicety. For students, researchers, and anyone producing work they'll be held accountable for, being able to audit the reasoning path — and cite the exact sources behind each claim — is what makes an AI tool usable in serious contexts.

7. The Catch: Watermark-Free Exports Still Need Ultra

For all this new power, one thing did not change: free and Plus exports still carry a "Made with NotebookLM" watermark. Removing it officially still requires NotebookLM Ultra, which sits at the top of Google's pricing ladder. If you're weighing whether that's worth it, our Free vs Plus vs Ultra comparison breaks down exactly what each tier gets you.

So the update makes NotebookLM vastly more capable, but the monetization lever stays firmly in place. More formats, same watermark policy.

What This Means for Watermark Removal

Here's the practical upshot. Before this update, NotebookLM watermarks showed up on a limited set of outputs — mostly PDF slides, videos, and infographics. Now the surface area has exploded:

• Editable PPTX presentations — watermarked.
• DOCX reports — watermarked.
• XLSX and CSV data exports — watermarked (or watermark-adjacent branding).
• Generated charts and visualizations — watermarked.
• Videos, PDFs, infographics, audio — still watermarked, as before.

More export formats means more watermarks, which means more people needing a clean way to strip them. A student exporting a DOCX study guide, a consultant building a PPTX deck, an analyst pulling an XLSX dataset — all of them now hit the same "Made with NotebookLM" tax on professional deliverables.

That's exactly what our tool is for. It runs entirely in your browser — files never touch a server — and handles every NotebookLM export format: video (FFmpeg WASM delogo), PDF and slides (connected-component detection plus gradient fill), PPTX (unpack, clean, repack), infographics, and Gemini images (lossless alpha-channel reversal). As Google keeps expanding the export menu, a free, private, format-agnostic remover only becomes more useful.

The Takeaway

The June 8, 2026 agentic update is a genuine leap. NotebookLM went from a passive research reader to an active agent that runs code, researches from scratch, and exports editable files across nearly every office format — all powered by Gemini 3.5 and the Antigravity framework, with transparent reasoning you can actually audit.

It's a stronger tool by every measure. But the watermark on free and Plus exports is still there, now spread across far more formats than before. If you're producing anything professional, the last step in your workflow is the same as it always was — clean the export. The difference is you'll be doing it a lot more often.

Frequently Asked Questions

Do the new PPTX, DOCX, and XLSX exports have watermarks?

Yes. On the free and Plus tiers, the expanded export formats carry the same "Made with NotebookLM" watermark branding as the older PDF and video exports. Only NotebookLM Ultra removes it officially — or you can strip it for free with a browser-based tool after exporting.

Does the agentic update change how watermark removal works?

Not fundamentally. The watermark is still applied to the exported file, so it can still be removed after export. The main change is that there are now more file types to clean — editable PPTX, DOCX, XLSX, CSV, and charts — in addition to the existing video, PDF, slides, and infographic formats. A format-agnostic remover handles all of them.

Is the start-from-scratch research mode safe to rely on for accuracy?

It's more auditable than typical AI web search because NotebookLM proposes sources for you to approve and shows its chain-of-thought reasoning, so every claim traces back to a visible source. Still, treat AI-discovered sources the way you'd treat any search results — review them before generating final outputs, especially for academic or professional work.