NotebookLM 3.5: From Preprocessor to Agentic Research Partner

What Google announced on June 8, 2026

The largest NotebookLM update so far moves the product out of the "smart reader for uploaded sources" corner and toward being an agentic research partner. Three building blocks stand out:

1. Gemini 3.5 as the new default model — with visible "thinking steps" that show how an answer was assembled.
2. Secure cloud computer — NotebookLM can now research the web on its own, collect sources and drop them into the notebook.
3. Antigravity integration — multi-step agent workflows directly inside the notebook, instead of plain Q&A over existing sources.

On top of that: more file formats, new output formats, and a chat-first entry point that builds an entire notebook from a single question (TechCrunch).

The new features in detail

1. Gemini 3.5 with visible reasoning

NotebookLM now exposes its reasoning steps — similar to Claude or DeepSeek R1. That's more than cosmetic: researchers can finally trace which source contributed to which argument, and where the model interprets rather than quotes.

2. Secure Cloud Computer: NotebookLM researches on its own

Possibly the biggest leap. The old rule — "NotebookLM only knows what you feed it" — is gone. With the new cloud computer the notebook opens a browser in a secure sandbox, pulls pages, extracts content and drops it into your sources list.

In effect, it's a deep research agent — comparable to OpenAI Deep Research, Perplexity Pro Search or Claude Computer Use, but cleanly wired into NotebookLM's source / citation / audio-overview logic.

3. Antigravity workflows

The Antigravity integration enables multi-step agent pipelines: "Find every press release from competitor X in Q2, extract product announcements, summarize per month, export as a mind map." This shifts the work from prompt engineering to workflow engineering inside the notebook.

4. More formats — in and out

• Input: additional code and office formats, larger source quotas depending on plan
• Output: beyond audio overviews and mind maps — structured exports and new visualizations
• Build from chat: a single prompt assembles a full notebook including sources (Ars Technica)

Use cases: what actually changes

Competitive intelligence

Before: you collect sources, NotebookLM summarizes. Now: you say "watch these 6 competitors weekly," and NotebookLM fetches the sources itself.

Research & editorial

A research question becomes — in one step — a notebook with cited sources, mind map and audio overview. For editorial teams, setup time per topic drops from hours to minutes.

Team knowledge bases

"Build from chat" plus the cloud computer turn NotebookLM into a living knowledge base: one notebook per product line, customer or market, kept current on its own.

Honest learning

Visible reasoning steps make NotebookLM a more honest study partner: you see where the model is confident, where it interprets, where sources are thin.

What this means for our NotebookLM + Claude Code workflow

In our article NotebookLM + Claude Code we framed NotebookLM as a clean preprocessing layer in front of Claude Code: NotebookLM transcribes and indexes, Claude Code produces the output. NotebookLM 3.5 challenges that split in two places:

• Source acquisition: NotebookLM no longer needs you as a "source curator." The cloud computer takes over.
• Structured outputs: through Antigravity, NotebookLM moves into Claude Code's territory.

What stays: for code-heavy pipelines, repo operations and productive output layers, Claude Code is still superior. NotebookLM remains the stronger tool for source work with citation integrity. The combination doesn't become obsolete — it just needs rebalancing.

Critical take

• Privacy stays the main issue — "secure" cloud computer or not. For EU customers with sensitive data, the vendor risk barely moves.
• Vendor lock-in intensifies: the more notebooks collect sources and encode workflows on their own, the more expensive switching becomes.
• Availability: rollout starts on AI Ultra; broader access follows later.
• "Less like itself": early voices (MakeUseOf) warn that NotebookLM is blurring into the Gemini app with every expansion.

Bottom line

NotebookLM 3.5 isn't an iteration — it's a role change. The calm source reader becomes an agentic research partner with its own browser and workflow engine. For AI-first teams that means: re-evaluate NotebookLM — no longer just a preprocessor, but a serious deep-research layer next to Claude, Perplexity and OpenAI.

Sources

• Google Blog – Do better research with NotebookLM (June 8, 2026)
• The Verge – Gemini 3.5 upgrade & cloud computer
• Ars Technica – Gemini 3.5 and Antigravity come to NotebookLM
• TechCrunch – Build source repository from chat
• SiliconANGLE – Coding features & Gemini 3.5
• Google Blog – Notebooks in Gemini (April 2026)

→ Book an AI workflow workshop