- Published on
NotebookLM 2.0, Claude Code Dynamic Workflows & Gemini 3 Flash: July AI Tool Briefing
In July, the AI tool landscape shifted again β dramatically.
NotebookLM is no longer just a note-taking tool. It now has an embedded cloud computer that runs code, generates charts, and auto-creates PowerPoint presentations. Claude Code officially introduced Dynamic Workflows β a framework where AI agents orchestrate other AI agents in parallel β opening a new frontier for complex software engineering automation. And Google closed the Gemini 2.5 chapter, promoting Gemini 3 Flash as its default model. As an EdTech CEO and AI tool researcher, here is why these three changes matter right now.
Table of Contents
- NotebookLM 2.0: A Cloud Computer Inside Every Notebook
- Claude Code Dynamic Workflows: Multi-Agent Orchestration Goes Official
- Gemini 3 Flash: The New Default That Replaced 2.5
- The Signal All Three Are Sending
1. NotebookLM 2.0: A Cloud Computer Inside Every Notebook
On June 8, 2026, NotebookLM received its biggest upgrade ever β transforming from a document analysis tool into a research execution engine.
NotebookLM was originally built around a simple premise: upload your materials, ask questions, and get answers that stay grounded in those materials. That grounding prevented hallucination and made it ideal for analyzing papers or reports. This update rewrites that definition entirely.
![NotebookLM 2.0 β the cloud computer running a data analysis request. The chat panel shows Python execution output and a generated chart side by side.]
Every Notebook Gets Its Own Cloud Computer
The centerpiece of this update is the embedded cloud computer. Each NotebookLM notebook now comes with its own cloud computer that can write and run code in the background while you converse in natural language.
Type "Show me a line chart of monthly trends from this dataset" and the AI writes Python code, executes it, and returns the chart β all inside the chat window. Over 100 pre-configured software skills are built in, covering statistical analysis, data visualization, and text processing. Complex analytical tasks that previously required switching to a separate coding environment now happen in a single conversation.
Gemini 3.5 Upgrade: You Can Now See the Thinking
NotebookLM's underlying model has been upgraded to Gemini 3.5. The most significant consequence is not raw capability β it is visible reasoning.
Previously, NotebookLM showed only the final answer. Now, when AI works through a complex question, it shows each reasoning step in the chat window: which sources it referenced, what assumptions it made, and how it arrived at its conclusion. For research-heavy work, this means you can verify the AI's logic rather than simply trust the output.
Web Research Integration and Expanded Export Options
NotebookLM used to be strictly bounded by what you uploaded. The new web research capability lets you start with a loose idea and have the AI find and organize relevant web sources to build your research repository automatically.
Export options have also expanded significantly. Beyond the existing Audio Overview (podcast format), NotebookLM 2.0 now generates editable PDFs and PowerPoint presentations that integrate directly with Google Slides.
| Feature | Previous NotebookLM | NotebookLM 2.0 |
|---|---|---|
| AI Model | Gemini 2.5 | Gemini 3.5 |
| Code Execution | Not available | Built-in cloud computer |
| Data Sources | Uploaded files only | Uploaded files + web research |
| Output Formats | Text, audio | Text, audio, PDF, PPTX, charts |
| Reasoning Visibility | Not available | Step-by-step reasoning shown |
Access Limitation: Currently Behind a Paywall
The cloud computer, Gemini 3.5, and web research features are currently available only to Google AI Ultra subscribers and Workspace business customers. Free users do not yet have access.
EdTech Perspective: NotebookLM 2.0 could transform how educators approach research, curriculum development, and school data analysis. A lesson plan where students upload a dataset, ask questions in natural language, and watch the AI generate charts and analysis β that is now possible. Verify subscription costs and data privacy policies before bringing this into a classroom.
2. Claude Code Dynamic Workflows: Multi-Agent Orchestration Goes Official
In June 2026, Anthropic formally introduced Dynamic Workflows to Claude Code β a capability that coordinates large numbers of AI agents to tackle complex software engineering tasks that no single agent can handle alone.
Before Dynamic Workflows, Claude Code operated linearly: one AI, one task, one thread. Dynamic Workflows changes this entirely. Claude now dynamically creates orchestration scripts, breaks work into subtasks, runs them in parallel across multiple subagents, validates results, and presents a synthesized final answer.
![Claude Code Dynamic Workflows β the /workflows dashboard showing an orchestrator agent running five subagents in parallel, with real-time status and token usage for each]
Why Dynamic Workflows Were Needed
Even with Claude Sonnet 5's 1M-token context window, single-agent architectures hit practical limits on large codebases. Analyzing every file simultaneously, refactoring dozens of modules while tracking cross-file dependencies, or auditing a sprawling system for security vulnerabilities β these tasks exceed what a single context window can reliably hold.
Dynamic Workflows solve this with a divide-and-conquer approach. Large tasks are split into subtasks, each subagent handles its own bounded scope, and an orchestrator integrates the results.
[Orchestrator Agent]
β Decomposes task and assigns work
ββ [Subagent 1] Security vulnerability scan (auth/ module)
ββ [Subagent 2] Performance analysis (database/ module)
ββ [Subagent 3] Test coverage audit (tests/)
ββ [Subagent 4] Documentation update (docs/)
β Consolidates results β Final report
Real-World Use Case: Large-Scale Code Migration
The scenario where Dynamic Workflows delivers the most value is large-scale code migration. Asking a single agent to migrate 100 files from Python 2 to Python 3 frequently results in context overflow or mid-task coherence failures.
Dynamic Workflows automatically groups files, assigns them to agents, resolves inter-file dependencies, and has a verification agent cross-check each migration. Reported use cases include large-scale bug investigation, legacy code migration, security audits, performance reviews, and architecture-wide analysis.
Sonnet 5 as Default: 1M Tokens + Promotional Pricing
Alongside this update, Claude Sonnet 5 became the default model in Claude Code. Sonnet 5 has a native 1M-token context window and runs at promotional pricing (10 per Mtok) through August 31, 2026.
Agent Teams Simplified: TeamCreate/TeamDelete Retired
The June 15 update retired the TeamCreate and TeamDelete tools. Setting CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 now creates an implicit team for every session β spawn teammates directly via the Agent tool's name parameter, no setup step required.
EdTech Perspective: Dynamic Workflows can be applied directly to large-scale educational content production. Designing learning objectives, activities, and assessments for 40 lessons in parallel β what would take a curriculum team weeks β can be drafted in a day. Final review by a subject-matter expert remains essential; AI produces the scaffolding, not the judgment.
3. Gemini 3 Flash: The New Default That Replaced 2.5
Google closed the Gemini 2.5 era and elevated Gemini 3 Flash as its default model in the Gemini app, with Gemini 3.5 Flash positioned as the flagship for demanding tasks.
For over a year, the Gemini 2.5 series (Flash and Pro) was Google's primary lineup. Now Google is transitioning to the third generation across the board.
![Gemini 3 Flash β Google AI Studio model selector showing Gemini 3 Flash and 3.5 Flash options, with a complex math problem being solved step by step]
Gemini 3 Flash: The New Everyday Standard
Gemini 3 Flash is now the default model in the Gemini app. Google describes it as a "major capability upgrade" over Gemini 2.5 Flash, designed for everyday tasks and multi-step projects at high speed.
Gemini 3.5 Flash: The Flagship for Complex Work
Gemini 3.5 Flash is positioned as the best model for "getting challenging tasks done quickly and efficiently." It is specifically designed to navigate real-world complexity β the kind of multi-step reasoning and domain-crossing analysis that requires sustained coherence across a long chain of thought.
Deep Research: File Uploads Now Supported
Google added a significant capability to Deep Research: users can now upload files and images directly as sources for research reports. This allows combining proprietary data with public web information in a single AI-generated research document.
Deep Research reports can now also be converted into interactive visuals and quizzes via Canvas. The Deep Research feature itself is available on Gemini 2.5 Flash at no cost for all users.
| Model | Position | Key Strength |
|---|---|---|
| Gemini 3 Flash | Default in Gemini app | Everyday tasks, fast response |
| Gemini 3.5 Flash | Flagship for complex work | Deep reasoning, multi-step projects |
| Gemini 2.5 Pro | Enterprise GA (maintained) | Scientific discovery, legacy code migration |
EdTech Perspective: The Deep Research file upload feature has direct classroom applications. Upload internal school data (academic results, survey responses) alongside a web research query, and AI can generate a contextual educational analysis report. Always verify FERPA / data privacy compliance before uploading student data.
4. The Signal All Three Are Sending
NotebookLM 2.0, Claude Code Dynamic Workflows, Gemini 3 Flash. Three updates from different companies targeting different use cases β but pointing in the same direction.
"AI now shows you how it thinks, not just what it concludes."
NotebookLM surfaces its reasoning steps in the chat. Claude Code's /workflows dashboard shows every agent working in real time. Gemini 3.5 Flash emphasizes step-by-step problem-solving as a core differentiator.
AI is evolving from a black box that produces answers into a transparent collaborative partner that shows its work. In education, this matters enormously. Students who can watch AI reason through a problem β and then critique that reasoning β are learning a skill that will define competence in the AI era.
Three Things to Try Right Now
NotebookLM 2.0 (Google AI Ultra subscribers): Open a notebook, upload a CSV data file as a source, then type: "Analyze this data and build a monthly trend chart." Watching the cloud computer execute code and return a chart in the chat window will make the before/after contrast immediately obvious.
Claude Code Dynamic Workflows: Run claude then enter /workflows to open the dashboard. Start with a small task β "Find all TODO comments in this project and organize them by priority" β to observe how the workflow distributes work across agents. Then scale up.
Gemini 3.5 Flash + Deep Research: In Google AI Studio, switch to Gemini 3.5 Flash and run a Deep Research session. Upload 2β3 relevant PDFs and request a web-inclusive research report on the same topic. The final output β your materials plus live web sources, synthesized β can be converted to a Canvas interactive presentation immediately.
If one of these three updates is something you want to apply to your work first, share your use case in the comments. Real-world scenarios from practitioners make the best discussion.
Related Reading
- Claude Code 5-Level Subagents, OpenClaw & Suno Voice Cloning: July AI Tool Briefing
- Claude Sonnet 5 as Default, Notion Workers & Cursor $60B Acquisition: July 1 AI Briefing
- Claude's Three Big Innovations in Late June 2026: Dynamic Workflows, Opus 4.8 & Slack Agent
Sources
- Google Blog (2026, Jun 8). Do your best research with NotebookLM. https://blog.google/innovation-and-ai/products/notebooklm/better-research-notebooklm/
- Google Workspace Updates (2026, Mar). New ways to customize and interact with your content in NotebookLM. https://workspaceupdates.googleblog.com/2026/03/new-ways-to-customize-and-interact-with-your-content-in-NotebookLM.html
- Geeky Gadgets (2026). NotebookLM 2.0: A Complete Guide to the 2026 Update. https://www.geeky-gadgets.com/notebooklm-2026-new-features/
- InfoQ (2026, Jun). Claude Code Adds Dynamic Workflows for Parallel Agent Coordination. https://www.infoq.com/news/2026/06/dynamic-workflows-claude-code/
- Claude Code Docs (2026, Jun). Claude Code Changelog. https://code.claude.com/docs/en/changelog
- SitePoint (2026, Jun). Claude Code June 2026: 10 New Features Devs Need to Know. https://www.sitepoint.com/claude-code-june-2026-10-new-features-devs-need-to-know/
- Google Developers Blog (2026). Gemini 2.5: Updates to our family of thinking models. https://developers.googleblog.com/en/gemini-2-5-thinking-model-updates/
- Google Cloud Blog (2026). Gemini 2.5 Flash-Lite, Flash, Pro GA on Vertex AI. https://cloud.google.com/blog/products/ai-machine-learning/gemini-2-5-flash-lite-flash-pro-ga-vertex-ai