Published on

AI Tool Digest June 2026 β€” NotebookLM Agentic Research, Claude Code Multi-Agent, Gemini Omni Video Generation

The AI ecosystem is finally outgrowing the era of "tools you use alone."

Three major updates this month all point in the same direction: AI that plans on its own, executes on its own, and collaborates with other AI agents without waiting to be asked.

As an EdTech CEO and daily AI practitioner, here are the three changes that matter most right now β€” framed through the lens of what you can actually do with them.

Table of Contents

  1. NotebookLM 2.0 β€” "Now the AI Goes Out and Finds the Sources"
  2. Claude Code Multi-Agent β€” "Agents That Spawn Agents"
  3. Gemini Omni β€” "The Video Creation Revolution Through Conversation"
  4. What These Three Updates Are Pointing Toward

1. NotebookLM 2.0 β€” "Now the AI Goes Out and Finds the Sources"

NotebookLM Agentic Research Interface Placeholder

On June 8, 2026, Google announced the biggest upgrade in NotebookLM's history.

The old NotebookLM analyzed documents you uploaded to it. That's changed. Now you throw out a topic, and NotebookLM runs its own Google searches, finds and suggests relevant sources, executes code to analyze data, and shows you its reasoning step by step.

Agentic Research: Start With Nothing

Before, opening NotebookLM meant having a stack of PDFs or links ready in advance. That sequence is reversed now.

Start a conversation about your project topic, and NotebookLM uses Google Search to proactively suggest relevant sources and build your knowledge base for you. The starting point can be "I'm not sure where to begin."

Why is this significant? Until now, AI only processed information you provided. NotebookLM can now go out and collect it.

Code Execution: A Computer Inside Your Notebook

NotebookLM Code Execution Placeholder

Powered by the Gemini 3.5 model, each notebook now includes a secure, isolated cloud computer for running code.

  • Execute Python directly and analyze data
  • Generate charts, spreadsheets, PDFs, and PowerPoint files
  • See the chain-of-thought reasoning behind every conclusion

Imagine uploading a set of research papers and saying "compare these datasets and visualize the differences." NotebookLM writes the code, runs it, and hands you the charts.

Applications in educational settings:

ScenarioHow It Works
Student research guidanceAI helps collect sources from a topic alone
Data analysis classesLive visualization through code execution
Lesson material creationGenerate PDF, PPT, and spreadsheets simultaneously
Paper reviewsTransparent reasoning enables real verification

Showing Its Reasoning: The Condition for Trusting AI

The most important feature in this update, to my mind, is the visible reasoning chain.

The biggest reason people distrust AI outputs is that the path to a conclusion is invisible. NotebookLM now doesn't just show you results β€” it unfolds the reasoning step by step. If something looks wrong at a particular step, you can point to it and ask for a correction.

This is the condition that had to be met before AI could be a genuine research partner.

EdTech CEO perspective: NotebookLM has become an all-in-one research partner that handles searching, executing, and reasoning. How students learn to do research may fundamentally change. Teaching "how do you verify it?" matters more now than "where do you look?"

2. Claude Code Multi-Agent β€” "Agents That Spawn Agents"

Claude Code Agent View Placeholder

The core of Claude Code's June update is one thing: AI agents can now create and delegate to other agents.

In developer circles, "vibe coding is passΓ©" has become a common phrase. In February, Andrej Karpathy declared vibe coding outdated and proposed a more structured agentic engineering paradigm. Claude Code is at the center of that shift.

Agent View: Manage Multiple AIs in One Screen

You can now manage multiple Claude sessions simultaneously from a single CLI view.

# Claude Code Agent View example
● Session 1: [Reviewing code]     ← active work
● Session 2: [Running tests]      ← background
● Session 3: [Writing docs: waiting] ← needs input
  • Manage multiple agents at once: start, background, peek at status, jump back in
  • Only intervene when needed: step in only when an agent is stuck
  • Workflow visibility: see each agent's last response instantly

Think of it as becoming a team lead whose team members are agents. They handle their own work; you unblock them when they need it.

Sub-Agents Spawning Sub-Agents: Five Levels Deep

The technical core of this update is hierarchical agent architecture.

Claude Code agents can now delegate tasks they can't efficiently handle to other agents. Those sub-agents can spin up their own sub-agents. The structure supports up to five levels of depth.

LevelExample Role
Level 1 (Orchestrator)Overall project planning and coordination
Level 2 (Specialist)Code writing, documentation, testing
Level 3 (Sub-task)File parsing, API calls, validation
Levels 4–5Fine-grained parallel processing

A single prompt can now trigger dynamic workflows where dozens to hundreds of sub-agents process work in parallel.

/cd Command and Auto Mode

Two practical additions worth noting.

/cd command: Move the current session to a different working directory without rebuilding the prompt cache. Context is preserved across the directory change β€” useful when continuing a long-context task across multiple repositories.

Auto mode (research preview): An AI classifier handles permission prompts automatically. Safe operations proceed without interruption; risky ones are blocked for human review. No more answering "Allow this?" on every step.

Rate limits doubled: Anthropic doubled Claude Code's API rate limits, making it significantly more reliable for sustained development work.

EdTech CEO perspective: With 46% of new code being AI-generated, this architecture is more than a developer tool. Automating student code feedback or generating multi-stage curriculum content becomes achievable. The framing shifts from "using AI" to "running an AI team."

3. Gemini Omni β€” "The Video Creation Revolution Through Conversation"

Gemini Omni Video Generation Placeholder

Gemini Omni, unveiled at Google I/O 2026, is eliminating the barrier to entry for video creation.

Previous AI video tools required precise prompts. If the output didn't match your vision, you started over from scratch. Gemini Omni flips this model.

Edit Video Through Conversation

The defining feature of Gemini Omni is conversational editing.

  • "Change the background to an evening cityscape."
  • "Add a slow zoom-in to this scene."
  • "Make the lighting consistent with the previous scene."

Natural language instructions like these directly modify the video. You can mix any inputs β€” text, images, audio, existing video. Because previous instructions are remembered, characters, lighting, and objects stay consistent across edits.

An AI That Understands Physics

Gemini Omni Physics Simulation Placeholder

What separates Gemini Omni technically from earlier video AI is its understanding of physical laws.

The model understands phenomena like gravity, kinetic energy, and fluid dynamics, so water flows naturally and objects behave plausibly in generated video. The "physically wrong" moments common in earlier AI video are greatly reduced.

Gemini's knowledge of history, science, and culture is also woven in. Ask for "a Joseon Dynasty New Year scene" and the generated visuals can reflect that historical knowledge.

Access and Availability

ItemDetails
Current modelGemini Omni Flash (10-second video generation)
Access pointsGemini app, Google Flow
Subscription requiredGoogle AI Plus, Pro, or Ultra
YouTube integrationYouTube Shorts Remix, YouTube Create (18+, free)
WatermarkSynthID digital watermark embedded automatically

The free access through YouTube Shorts Remix stands out. Over two million YouTube creators can now reach an AI video tool without a subscription.

EdTech CEO perspective: The barrier to producing educational video content is dropping fast. Teachers making their own lesson videos or students adding video to presentations is no longer a specialized skill. At the same time, the importance of transparency and media literacy education β€” "who made this and how?" β€” rises equally.

What These Three Updates Are Pointing Toward

The word threading through all three June updates is agency.

NotebookLM finds sources and runs code before you even ask. Claude Code builds a layered architecture where agents delegate to agents. Gemini Omni translates natural language intent into video in real time.

One common thread: AI is now handling the intermediate steps on its own.

What remains squarely in human hands is the judgment call β€” what to ask for, what to verify, what to trust. The gap between people who internalize this shift early and those who don't will widen faster from here.

Practical Tips

1. NotebookLM: Start a research session with no sources Open a new notebook, skip the upload step, and type your project topic directly into chat. Watch what sources it suggests, and use it as a partner that helps you find your research direction.

2. Claude Code: Delegate a task to a sub-agent Pick one routine task and hand it off using the /agent command. While it works in the background, take on something else β€” experience firsthand how multitasking changes when the other worker is an agent.

3. Gemini Omni: Try YouTube Shorts Remix YouTube Shorts Remix lets you access Gemini Omni for free. Load an existing Short and try changing the background or lighting through natural conversation.

4. Build a habit of checking reasoning steps When NotebookLM shows its chain of thought, don't just use the final output β€” scan the intermediate steps once. It's the fastest way to build calibrated trust in AI results.

AI tools maturing means they become more powerful β€” but it also means more judgment calls fall to you. Which agent gets which task, which output gets verified, which result gets trusted: those decisions remain human territory.

Related Posts

Sources:

Discuss on Twitter β€’ View on GitHub
AI Tool Digest June 2026 β€” NotebookLM Agentic Research, Claude Code Multi-Agent, Gemini Omni Video Generation | MINSSAM.COM