NotebookLM writes code and spreadsheets, backed by Gemini 3.5
NotebookLM
Google
TLDR: WASHINGTON—Google upgraded NotebookLM with Gemini 3.5 so it can write and execute code, export files, and source research from prompts.
Key Takeaways:
- NotebookLM now runs on dedicated cloud computers with Gemini 3.5 and Antigravity coding support.
- It can generate PDFs, Word, Excel, PowerPoint, CSV, data visualizations, and images, plus execute code.
- Google benchmarks claim a 65 percent win rate improvement, with users keeping control of attributed sources.
- Research starts from a rough question using Google Search, while NotebookLM builds a repository users approve.
This is NotebookLM inching from helper to doer, where the notebook becomes a mini lab. The real unlock is control plus attribution, so you can move fast without losing the trail.
This is NotebookLM inching from helper to doer, where the notebook becomes a mini lab. The real unlock is control plus attribution, so you can move fast without losing the trail.
Q&A
What changes for teams when a notebook can both draft and execute code in the same workflow?
Work shifts from pasting scripts into separate tools to iterating inside one shared notebook, reducing handoffs and speeding debugging cycles.
Why does Google emphasize transparency in how the AI reaches conclusions?
As notebooks produce code, charts, and citations, traceable reasoning helps users audit outputs and defend decisions to reviewers.
What happens when users can start research from a rough idea instead of a prepared source set?
The bottleneck moves from sourcing to judgment, forcing teams to spend more time approving sources and less time assembling libraries.
How might the downloadable formats change how people share work with non technical teammates?
Exporting Excel, PowerPoint, and PDFs turns notebooks into shareable deliverables, so collaborators can consume results without adopting the tool.
Could dedicated cloud execution increase costs or introduce new limits for heavy spreadsheet and code workloads?
If compute is centralized, the platform may meter or constrain intensive runs, making performance and quota planning part of adoption.
Top in Tech Trends
AGILINK reframes dexterity as the contact problem at ICRA 2026
Computer Science Graduates Face Hiring Mismatch, Not a Dead Degree
Continuing education fights AI job churn and tech acceleration
Formlabs pushes Fuse X1 to make industrial SLS attainable
No comments yet. Be the first to share your thoughts!