NotebookLM Gains Cloud Environments and Gemini 3.5 Agents

On June 8, 2026, Google released a comprehensive update to NotebookLM that transitions the tool from a static note-taking application into an autonomous research assistant. The platform is now powered by the Gemini 3.5 model family, utilizing Gemini 3.5 Flash for long-horizon reasoning. This release marks the integration of Google’s new Antigravity agent-first development architecture, enabling proactive document analysis and multi-step research workflows.

Architecture and Cloud Execution

Every notebook now provisions a dedicated secure cloud computing environment upon creation. This infrastructure allows NotebookLM to write and execute code natively to perform complex data analysis directly within the platform. The Antigravity architecture decouples the reasoning engine from the execution environment, providing better visibility into the AI’s decision-making process.

The updated system ships with over 100 curated agent skills pre-loaded. These capabilities allow the model to autonomously map deep connections across uploaded documents without requiring precise user prompting for every analytical step, operating much like standalone AI agents.

NotebookLM also introduces native Google Search integration for source discovery. Unlike earlier iterations that strictly constrained answers to the provided corpus, the autonomous web search operates as an expansion tool. The agent determines when internal documents lack sufficient detail and initiates external queries to find and attribute high-quality external sources based on initial user prompts.

Benchmark Results

Google evaluated the upgraded architecture against the previous NotebookLM baseline across internal testing metrics. The integration of Gemini 3.5 and Antigravity produced measurable gains in advanced research tasks.

Evaluation Category	Win Rate vs Previous Baseline
Source Discovery	78.2%
Large Document Analysis	69.9%
Core Dimensions (Average)	> 65.0%

Artifact Generation

The output layer now supports assembling research context into downloadable artifacts. The system generates structured data formats alongside standard document outputs, relying on strict schema adherence to populate complex files reliably.

Supported export formats include:

• Documents: PDF, DOCX, Markdown, and plain text.
• Data and Visuals: CSV, JSON, and XLSX for structured data, plus PNG and SVG formats for data visualization.
• Presentations: PPTX slide decks.
• Multimedia: Image generation powered by Nano Banana 2, supporting PNG, JPG, and GIF exports.

Availability

The updated NotebookLM is currently rolling out globally. Access is restricted to Google AI Ultra subscribers and Google Workspace business customers on the “AI Expanded” tier. This rollout follows the foundational strategy laid out during Google I/O 2026 in May, where the company detailed a broader shift toward persistent, agent-based AI systems.

For developers building enterprise knowledge applications, the native integration of secure code execution and web grounding within a single managed workspace establishes a new baseline. You should evaluate how these embedded agent capabilities alter the necessity of maintaining custom retrieval and execution pipelines for internal research tasks.