NotebookLM Becomes Gemini Notebook, Claude Code Agents Go 3 Levels Deep, and Suno Enters iMessage: 3 AI Picks for Late July 2026

In the second half of July 2026, three major AI tools transformed again.

Google renamed a product used by 30 million people. It was not a simple rebrand — a cloud computer now lives inside every notebook, letting users write and run code directly in their research space. Anthropic changed how Claude Code subagents work: they can now create their own subagents, forming a three-level agent hierarchy. And Suno AI rebuilt its most-criticized feature — the lyrics editor — while also embedding itself inside the iOS iMessage keyboard.

All three updates share one theme. Each one identified a specific weakness users had complained about — and fixed it directly.

Gemini Notebook: NotebookLM Starts Running Code
What the Cloud Computer Inside Your Notebook Actually Does
Claude Code Nested Subagents: Agents That Spawn Agents
The 20-Agent Concurrency Cap and What Depth 3 Means
Suno AI Lyrics Editor Rebuild: Refine Lyrics Like a Songwriter
Suno Moves Into iMessage
What These Three Updates Point To

1. Gemini Notebook: NotebookLM Starts Running Code

On July 16, 2026, Google renamed NotebookLM to Gemini Notebook. The functional change outweighs the branding one. Every notebook now has its own secure cloud computer, where you can write and execute code without leaving the app.

Since its 2023 launch, NotebookLM was known primarily as a tool for reading, summarizing, and answering questions about documents. Researchers, students, and educators uploaded papers, books, and lecture notes and conversed with the AI. But one thing remained out of reach: actually computing or analyzing the data in those uploads, generating charts, or running code to verify results.

Gemini Notebook crosses that line. A cloud execution environment — internally codenamed 'Antigravity' — is now attached to every notebook. Upload a CSV and instead of just getting a summary, you can run Python code to produce statistics and visualizations without leaving the notebook.

What Changed

Name: NotebookLM → Gemini Notebook (existing notebooks, shared links, and bookmarks continue to work as before)
Model upgrade: Transitions to Gemini 3.5 + Antigravity. Already live for AI Ultra subscribers; rolling out to Pro subscribers soon
Code execution: A secure cloud computer is attached to each notebook for writing and running Python
Search integration: Notebooks will soon surface in Google AI Mode search results
Gemini app sync: Cross-device note access integrates with the Gemini app

Rollout Status

Feature	Who Gets It
Gemini 3.5 + Code Execution (Antigravity)	AI Ultra subscribers, Workspace business accounts
Code Execution (expanding)	AI Pro subscribers (within weeks)
Rebrand (Gemini Notebook)	All users

Tips for Educators and Researchers

Think about what it means for a research tool to become an analysis tool. Upload a student survey as a CSV and the AI calculates statistics and draws a chart for you. Drop in research data and run analyses that were previously impossible inside a notebook environment. A teacher who uploads a curriculum plan can now ask the AI to visualize the structure.

"If NotebookLM was a great reading partner, Gemini Notebook is a research assistant that also does the data analysis."

2. What the Cloud Computer Inside Your Notebook Actually Does

You do not need to become a data scientist now that code execution exists. The key point is that you describe the analysis in plain language — Gemini Notebook writes the code and runs it for you.

Type "Show me average scores by age group from this survey data as a bar chart" and Gemini generates Python, executes it, and displays the result. You can inspect or edit the code — but you do not have to. All you need is an intention.

Note that the feature is currently limited to AI Ultra subscribers and Workspace business accounts. AI Pro users are next in line, within a few weeks.

AI Ultra (personal): $249.99/month — full Gemini 3.5 + code execution available now
AI Pro (personal): $19.99/month — code execution pending rollout

3. Claude Code Nested Subagents: Agents That Spawn Agents

On July 24, 2026 (version 2.1.219), Anthropic re-enabled Claude Code subagents to spawn their own subagents. The default nesting depth is 3.

Previously, depth 1 — Claude spawning subagents — was the maximum. Nested subagents were temporarily disabled entirely on July 21 (version 2.1.217) and restored three days later at depth 3. Alongside this change, concurrent subagent execution was capped at 20.

How the Nested Structure Works in Practice

What does depth 3 actually look like?

Level 0: Orchestrator (the main Claude talking to you)
  ├── Level 1: Subagent A (codebase analysis)
  │     ├── Level 2: Subagent A-1 (frontend analysis)
  │     │     └── Level 3: Subagent A-1a (React component analysis)
  │     └── Level 2: Subagent A-2 (backend analysis)
  └── Level 1: Subagent B (report writing)
        └── Level 2: Subagent B-1 (section drafting)

Level 0 directs the whole task. Level 1 divides by domain. Levels 2 and 3 handle increasingly granular subtasks. Previously the hierarchy stopped at level 1.

Constraints to Know

Concurrency cap: No matter how deep the nesting, at most 20 subagents can run simultaneously
Depth is configurable: Set CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1 to revert to the old single-level behavior
Streaming visibility: The --forward-subagent-text flag now surfaces output from depth-2+ subagents in the stream

Where It Shines

Large-scale migrations: The top-level agent divides modules, each module agent spawns file-level subagents for parallel processing
Multi-stage quality verification: Code generation → test writing → bug verification, each handled by a separate agent layer
Complex research: Topic agent → subtopic agents → individual source agents in a hierarchical sweep

4. The 20-Agent Concurrency Cap and What Depth 3 Means

The 20-agent cap is a safety guardrail. It prevents an infinitely expanding agent hierarchy from producing unpredictable token consumption or cascading errors.

Anthropic's move to disable nesting entirely on July 21 and restore it at depth 3 three days later reads as a response to real-world incidents — cost spikes or misbehavior observed in production. Depth 3 with a 20-agent concurrency limit is the balance point between practicality and safety.

Practical guidance for typical users:

Small to mid-size projects (under 50 files): A single agent or depth-1 is sufficient
Large codebases (hundreds to thousands of files): Depth 2–3 delivers meaningful time savings
Cost management: Token consumption multiplies with each nesting level — start with the 'small' workflow size in /config and scale up as needed

5. Suno AI Lyrics Editor Rebuild: Refine Lyrics Like a Songwriter

On July 9, 2026, Suno AI completely rebuilt its most-criticized feature: the lyrics editor. What had been a simple text-editing box is now a full songwriting workspace.

The problems with the old editor were well-documented. Direct edits often produced results that clashed with the musical style. Preserving rhythm and rhyme while revising was difficult. Editing a single verse without disrupting the rest was frustrating.

Key Features of the New Editor

Feature	Description
Lyricist profile	Learns your lyric-writing style so AI suggestions stay consistent
Natural language editing	Direct the AI: "make verse 2 more emotional," "strengthen the rhyme scheme"
Variations & References	3–5 alternative versions of the same passage for A/B comparison
Full-screen mode	A distraction-free writing space for focused lyric work
Structure labels	Verse, chorus, bridge markers make song structure visible at a glance
Autosave	Edits are preserved even if you close the browser mid-session

The Lyricist feature is particularly notable: it learns patterns in how you write — your preferred vocabulary, phrasing, rhyme density — so that when you ask for a revision, the AI's suggestion doesn't drift away from your established voice.

Tips for Educators and Content Creators

Songs for class: Turn curriculum content into lyrics, then instruct: "simplify the vocabulary for elementary students"
YouTube or podcast intros: Describe your channel concept and Lyricist drafts an opening in a style consistent with your brand every time
Language learning content: Ask the AI to weave specific vocabulary or grammar patterns into the lyrics to build learning songs

6. Suno Moves Into iMessage

On July 15, 2026, Suno AI integrated into the iOS iMessage keyboard. You can open Suno, create a song, and send it — without ever leaving the chat.

Previously, sharing a Suno song required three steps: create in the Suno app, copy the link, paste it into a message. The iMessage integration collapses that into one. Recipients can play the song directly in iMessage with no additional app install. And if they like it, they can open Suno from the same keyboard and send a song back.

"Making a custom birthday song and sending it over iMessage is now a 30-second process."

This integration marks a strategic shift: Suno is expanding from a music creation tool to a communication tool. The premise is that music can carry emotion in conversations the same way a voice note or a GIF does — and it should live inside the same messaging surface where those already work.

7. What These Three Updates Point To

Read side by side, the three updates reveal a clear direction for AI tools in late July 2026.

First, AI tools are moving from "read and write" to "compute and execute." Gemini Notebook's code execution is the most concrete example: a research tool that previously only processed text can now run calculations and verify results. AI is crossing from the language layer into the computation layer.

Second, agent autonomy is deepening hierarchically. Claude Code's nested subagents are not just about speed. They change how complex work gets structured: the user sets a goal, and the AI assembles a layered team to achieve it. The direction of travel — from single agent to team to hierarchical organization — will continue.

Third, AI is moving outside its dedicated apps. Suno's iMessage integration is symbolic of a broader pattern. AI capabilities are being embedded into existing communication infrastructure — messaging apps, email, search results. The future is AI that meets you where you already are, without requiring a context switch.

The people who benefit most from this transition are those who use AI to analyze data, delegate complex work to agent hierarchies, and reach for AI features naturally inside everyday tools — without stopping to open a separate app.

Final Thought

Gemini Notebook added a calculator to your research space. Claude Code added depth to its agent hierarchy. Suno brought music creation into daily conversation.

All three updates fixed something users had been complaining about. Watching which friction points get addressed each quarter is itself a useful way to understand where AI is actually heading.

Related Posts

Which of these three updates do you think will change how you work the most?

Sources

Table of Contents