- Published on
Claude Code Subagents Now Go 3 Levels Deep, Google Vids Gets AI Avatars, Suno Rebuilt Its Lyrics Editor: July 25 AI Digest
This week's AI tool updates all point to a single keyword: the depth of automation.
Claude Code subagents can now spawn subagents, which can themselves spawn subagents β a three-tier structure that is now the default. Google Vids edits video with a single line of text and builds AI avatars from a selfie. Suno rebuilt its lyrics editor from scratch and attached an AI that remembers your writing style like a dedicated collaborator.
These updates live in different domains, but they share a direction. Tools are starting to learn your style and context, and are beginning to automate the repetitive steps that used to require constant human input.
Table of Contents
- Claude Code v2.1.217: Nested Subagents at Depth 3 and What That Means
- Emoji Autocomplete and Filesystem Isolation Control: Small Updates, Real Impact
- Google Vids + Gemini Omni: AI Avatars and Text-Based Video Editing
- Suno Lyrics Editor Rebuilt: Lyricist Remembers Your Style
- Duration Slider and iMessage: Suno Moves into Everyday Life
- The Common Direction All Three Updates Share
1. Claude Code v2.1.217: Nested Subagents at Depth 3 and What That Means
Previously, a Claude Code subagent could not create its own subagents. Now the default nesting depth is 3 β agents can spawn agents that spawn agents.
This structure is called nested subagents. Claude Code 2.1.211 introduced the capability; 2.1.217 locks the default depth at 3.
![Claude Code workflow view β an orchestrator agent branching into two subagents, each of which branches into further child agents, forming a three-level tree structure]
What This Changes in Practice
Depth 1 means "I tell Claude to write code." Depth 3 is different.
Consider a blog automation pipeline:
- Tier 1 agent: Selects today's topics and spawns three subagents
- Tier 2 agents: Each runs a research agent and a writing agent in parallel
- Tier 3 agents: Three translation agents handle Korean, English, and Japanese versions of the draft
This entire flow runs unattended, from a single orchestrator prompt to completion.
| Item | Before (Depth 1) | Now (Depth 3) |
|---|---|---|
| Who creates subagents | Only the user | Agents can create agents |
| Parallelism model | Simple task decomposition | Hierarchical pipeline |
| Best suited for | Single tasks | Complex multi-step automation |
Tip: Nested subagents shine for complex, multi-stage work. For simple code requests, a single agent is still faster. The depth-3 default pays off on instructions like "analyze this entire project, refactor it, and run the tests" β anything that naturally breaks into distinct, concurrent phases.
2. Emoji Autocomplete and Filesystem Isolation Control: Small Updates, Real Impact
The same release included two less-headline-grabbing changes that quietly improve the daily development experience.
Emoji Shortcode Autocomplete
Typing : in the Claude Code prompt input now triggers emoji shortcode autocomplete β the same interaction pattern as Slack or Notion. A small convergence of communication grammar between chat tools and the development environment.
Independent Filesystem Isolation Control
Added in Claude Code 2.1.216, filesystem isolation can now be toggled independently of other settings. Previously, enabling isolation mode also changed other options as a package. Selective isolation makes security-sensitive workflows easier to configure precisely.
Additionally, /code-review now runs as a background subagent, so requesting a code review no longer interrupts or fills the current conversation thread.
![Claude Code settings panel β filesystem isolation appears as a standalone toggle, separated from other options, with an "Isolated: ON" badge visible in the interface]
3. Google Vids + Gemini Omni: AI Avatars and Text-Based Video Editing
On July 16, 2026, Google integrated Gemini Omni into Google Vids, transforming it from a presentation-style video tool into an avatar-based video creation platform.
Two capabilities are now central to the product.
Personal AI Avatar Creation
A selfie and a voice sample are enough to generate a digital avatar that closely resembles you. That avatar then appears in videos, speaking text you provide as a prompt. Teachers and instructors can produce course videos without stepping in front of a camera each time.
Text-Based Video Editing
Existing videos can be edited with natural language commands: "remove the background noise," "change the color grade to a warmer tone," "make this scene more minimal." The AI interprets the instruction and applies it without requiring the user to open traditional editing software.
![Google Vids interface β on the left, a preview of the user's AI avatar delivering a video; on the right, a chat panel with text commands like "Remove background noise, Color correction: warm tone" entered as editing instructions]
Educational Applications
From an EdTech perspective, this update lowers the barrier to content creation by another step.
Lecture videos: Teachers who are uncomfortable on camera now have a realistic alternative β an avatar that looks and sounds like them delivers the lesson.
Content updates: Rather than re-recording an entire video, a specific explanation can be updated through a text command, keeping materials current without a full reshoot.
Multilingual content: Replacing an avatar's lip sync and audio with a different language version of the same script is now technically viable, pointing toward lower-cost multilingual production.
Tip: Google Vids is available to Google Workspace subscribers at no additional cost. Rollout began July 16 for Rapid Release domains; Scheduled Release domains gain access on August 5, 2026. Educational institutions should verify access through their Google Workspace for Education accounts.
4. Suno Lyrics Editor Rebuilt: Lyricist Remembers Your Style
On July 9, 2026, Suno rebuilt its lyrics editor from the ground up. The most significant addition is Lyricist β a profile that locks in your writing style.
The previous Suno workflow was straightforward: type lyrics directly or edit AI-generated ones. The new Lyrics Editor changes the underlying concept of how lyric writing works.
Lyricist Profile: AI Learns Your Writing Style
Lyricist is a profile that learns and stores your preferred rhyme patterns, frequently used words and phrases, and tonal approach. Once you set a lock, all subsequent AI suggestions align with that style.
For anyone working on multiple artist projects simultaneously, separate Lyricist profiles can maintain distinct styles per project.
Natural Language Editing
You can now select a specific line in the lyrics and issue a plain-language instruction: "make this part darker," "revise the third-verse chorus to rhyme with the first verse." The AI understands the full structure of the song and modifies only the targeted section in context.
![Suno Lyrics Editor β the full lyrics are displayed on the left with structural labels (Verse 1, Chorus, Bridge); on the right, a chat panel shows the natural language command "Make the bridge darker in tone" entered as an edit instruction]
Other Updates in the Rebuilt Editor
- Variations & References: Set a specific lyric structure as a reference template and generate variations from it
- Structure labels: Sections are automatically labeled (Verse, Pre-Chorus, Chorus, Bridge, Outro)
- Full-screen mode: A distraction-free writing space for focused lyric work
- Autosave: No more losing edits mid-session
5. Duration Slider and iMessage: Suno Moves into Everyday Life
Alongside the Lyrics Editor, Suno shipped two additional updates: a song length slider and iMessage keyboard integration.
Duration Slider
Added July 20, 2026 to the web Create form, the Duration slider lets you set the target song length before generating β available on the V5.5 model. Previously the AI decided the length automatically. Now you specify from the start whether you need a 30-second intro or a full three-minute track.
iMessage Keyboard Integration
From July 15, 2026, iOS users can access Suno directly from the iMessage keyboard. While composing a message, open the Suno keyboard, configure a short song, and send it inline. Recipients play it directly inside iMessage without leaving the app.
Birthday messages, encouragement, jokes, and everyday greetings can now arrive as short Suno-generated songs instead of text.
Tip: The Lyricist feature translates well to educational content creation. When producing a series of subject-specific learning songs, locking a Lyricist profile to "elementary school vocabulary, repeated chorus, bright and rhythmic style" means you never have to re-explain those constraints for every new song in the series.
6. The Common Direction All Three Updates Share
Claude Code's nested subagents, Google Vids' avatar and text editing, Suno's Lyricist profile. On the surface these look like updates to different tools solving different problems. But the pattern is the same.
All three move in the same direction: the tool learns the user's context and style, and handles the repetitive steps internally.
- Claude Code builds a structure where agents coordinate agents autonomously
- Google Vids remembers your face and voice to produce videos without repeated recordings
- Suno remembers your lyric style so you never have to re-explain it from scratch
AI tools have started learning "you." That is the most important shift happening in July 2026.
Quick Action Summary
Test nested subagents in Claude Code: Give a complex multi-step instruction (research β write β translate) as a single prompt. Version 2.1.211 and above applies depth-3 nesting by default.
Create your Google Vids avatar: If you have a Google Workspace account, open Google Vids and generate a personal avatar from a selfie and voice sample. It can substantially reduce the time cost of producing recurring lecture videos.
Set up a Lyricist profile in Suno: Open the Lyrics Editor on Suno's web app and create a Lyricist profile based on your writing style. Lock it, then compare the AI suggestions before and after β the difference is immediate.
Further Reading
- Claude Code's Default Model Switched to Sonnet 5
- Gemini 3.6 Flash and Claude Reflect: AI Analyzes Your Own Patterns
Sources
- Anthropic (2026). Claude Code Release Notes β What's New. https://code.claude.com/docs/en/whats-new
- Havoptic (2026). Claude Code v2.1.217 (Jul 21, 2026) β Every Release, Summarized. https://www.havoptic.com/tools/claude-code
- Releasebot (2026). Claude Code Updates by Anthropic β July 2026. https://releasebot.io/updates/anthropic/claude-code
- Google Workspace Blog (2026). Cast yourself in AI video clips using your personal avatar with Gemini Omni in Vids. https://workspaceupdates.googleblog.com/2026/07/cast-yourself-in-ai-video-clips-using-your-personal-avatar-with-Gemini-Omni-in-Vids.html
- Google Blog (2026). Google Vids gets powerful upgrades with Gemini Omni. https://blog.google/products-and-platforms/products/workspace/gemini-omni-personal-avatars/
- Neowin (2026). Google Vids AI video maker adds Gemini Omni support, here's how it will help. https://www.neowin.net/news/google-vids-ai-video-maker-adds-gemini-omni-support-heres-how-it-will-help/
- Suno (2026). Suno Release Notes β Lyrics Editor Rebuilt, July 9 2026. https://suno.com/release-notes
- Roo (2026). Suno's New Lyrics Editor Fixes the One Thing Reviewers Kept Complaining About. https://roo.beehiiv.com/p/suno-lyrics-editor-update-july-2026