Welcome back to the AI Edge (by Miles Deutscher) weekly newsletter!
Claude has changed more in the last few weeks than most people realize. Claude 2.0 is officially here.
You can now assign one task to a team of helpers, manage other Claude Code sessions from a single chat, work directly in Docs and Google Sheets, and many more features.
This week, we're walking through seven of the most useful changes, with prompts you can try yourself.
This week’s edition is sponsored by our partner, QVAC. QVAC (by Tether) is the ultimate open-source local AI ecosystem. Be sure to stick around to learn more.
IN TODAY'S READ
Table of Contents
After our main guide, we’ll dive into the Midweek Edge covering the latest AI news and Looking Ahead, where we compile our top AI research, trends, and workflows.
WEEKLY BUILD & WORKFLOW
Give Claude a Team of Helpers
You can now give Claude a bigger job and have multiple helpers work on different parts at once, in supported Claude Code workflows.
Instead of doing the research, writing, and planning one after another, Claude delegates those tasks and brings the results back to the main conversation.
Let's say you're launching a digital product. You need competitor research, 𝕏 posts, and a sales page.
Paste this into a Claude Code session that supports subagents:
I'm launching a $49 [PRODUCT] for [TARGET AUDIENCE].
Start three helpers in parallel.
Helper 1: Research five competitors, including their offers,
pricing, positioning, and original sources.
Helper 2: Draft five X posts for the launch using my product
description and writing style.
Helper 3: Draft a simple sales page covering the problem,
offer, benefits, and FAQs.
Give every helper the same product and audience context.
When they're done, combine the results, flag anything that
conflicts, and tell me what still needs my decision.
Don't publish anything.You can check the helpers' progress from the task panel where available. For an ongoing project, keep the product brief and approved source material in one shared place so each helper works from the same facts.
Manage Your Other Sessions in One Chat
Running several sessions is useful, but it can be somewhat difficult to keep track of all your workflows.
Claude Code on desktop can check supported sessions, read their progress, and pass along changes. It’s essentially a new project manager feature.
Try this:
Act as my project manager. Review the Claude Code sessions
you can access.
Give me three lists: Needs my decision, Finished, and Stuck.
Keep each item to one line, with the session name and next action.
Suggest a clean way to group and rename the sessions.
Ask before changing or archiving anything.
You can also pass along an important correction:
Our product is for beginners, not experienced users.
Tell the launch-post and sales-page sessions to revise their
drafts for beginners. Show me what changed.
This works with sessions Claude Code can access, not every Claude conversation in your account. The desktop app's session list and cross-session messaging have different scopes.
Use Side Chats + Split Screen
Ever wanted to ask Claude a quick question without interrupting the job it's doing?
Side Chat lets you do that. Open the tab menu at the top-left of Claude Code and choose Side Chat, or type /btw in your current session. It uses the main conversation's context, but your question doesn't get added to the main thread.
Split Screen solves a different problem: overseeing multiple sessions at once.
Right-click another chat in the Claude Code sidebar and choose Open in → Split view. You can also hold Option on Mac or Alt on Windows while clicking a session.
Now you can keep research on one side and writing on the other, without constantly switching back and forth. Supported panels can also open in separate windows.
Edit Docs & Google Sheets Directly
Claude Docs lets you work on a document beside the conversation instead of copying the answer into another editor.
Ask Claude to make a document, review it as it writes, make your own edits, and share or export the finished work. Claude Design and Slides offer related ways to work on visual material.
You can also work inside Google Sheets with Claude's Workspace add-on. The add-on and the connector are different setup routes, so choose the one available on your account and review the changes before using the sheet.
/Skill-Doctor
If you've installed a lot of Skills and plugins, some may be adding unnecessary overhead to your Claude Code sessions.
A new skill called/skill-doctor. Its entire job is to audit old, bloated Claude skills.
Then ask:
Based on this report, which unused Skills should I disable first?
Rank them by context cost, explain each in one line,
and don't change anything yet.
Customize Claude With Mods
Claude Code now supports Mods, small TypeScript extensions that let you tweak Claude itself.
For example, you could create a Mod that sends a desktop notification when a long task finishes, with a one-line summary of what Claude did.
Example Prompt:
Build a simple Mod that notifies me whenever a task finishes
after running for more than one minute.
Include a one-line summary and a way to jump back to the session.
Show me the code and explain what permissions it needs.
Ask before installing it.Mods can be packaged inside plugins, and they run with the same computer access as Claude Code itself.
Other Features Worth Trying
Advanced Agent Management: Multi-session workflows where sub-agents can run parallel tasks and report back to a central conversation thread.
Cloud Sessions: Tasks can run remotely on Anthropic's servers, allowing sessions to continue even if you close your desktop.
Claude Duet: Anthropic’s competitor to Dots, Muse & Grok Bot (coming soon).
Built-in Browser & Live Data: Features an integrated browser interface to interact with web applications and build live, auto-updating dashboards or charts.
Desktop Automation: Includes "Computer Use" capabilities, allowing Claude to type, click, and navigate across screen applications with user permission.
Tether Releases QVAC: Ultimate Local AI Ecosystem
Tether (the company behind the USDT stablecoin) has open-sourced QVAC, a toolkit that runs AI models locally on your own hardware. No cloud, no API keys, no data leaving your device.
It works across Linux, macOS, Windows, Android, and iOS, and covers a huge range of tasks:
→ Text generation and chat
→ Speech-to-text and text-to-speech
→ Translation and OCR (reading text from images)
→ Image, video and music generation
→ RAG (letting AI search your own documents)
The most interesting part: QVAC can share models directly between devices peer-to-peer, like BitTorrent but for AI.
Why it matters: Your data stays private, there's no subscription, and it works offline. If you're running sensitive business or client work through AI, this is worth a look.
Quick start (for the more technical): npm i @qvac/sdk
It's fully open source (Apache-2.0) and free to use.
MIDWEEK EDGE: Manually curated list of the most important weekly AI news updates
Weekly AI News Recap
→ Google Announces Gemini 4 Carbon. A model focused on long-running engineering work and cybersecurity, including finding and fixing vulnerabilities.
Not publicly released yet (only announced).
Google is back in the AI race.
→ Gemini Can Now Edit Files Inside Chat. Google added direct editing for PDFs, documents, spreadsheets, presentations, and more.
No need for PDF/document editor tools anymore.
→ Google Sends AI Data Center to Space. Project Suncatcher has received its first response, generated by a data-center-class TPU in orbit. Insane!
→ OpenAI Updates How Its Models Should Behave. A new revised Model Spec places more emphasis on following user intent, transparency, human oversight, and minimizing harm.
TL;DR: new safeguards.
→ Google Launches Genie 4. Genie 4 creates realistic, interactive 3D environments with sound, dialogue, and characters that respond to your actions.
→ OpenAI Releases GPT-6 Rosalind. OpenAI launched a specialized GPT-6 model for life sciences research, designed to help scientists with biological analysis, drug discovery, and complex research workflows.
LOOKING AHEAD: Manually curated list of useful AI tools, skills, workflows & more
Tools and Research Worth Your Time
→ The Five (5) Levels of Making Money With AI.
Sell your time
Leverage people
Sell products
Own distribution
Deploy capital
→ 25 Upcoming AI Careers That Could Make You a Millionaire By 2030
→ Anthropic’s own internal use of Claude grew more than 250x from January to September (crazy!)
→ A New Way to Measure Useful AI.
OpenAI's LifeVal benchmark tests whether agents can help people complete everyday tasks, from managing a budget to finding childcare or handling paperwork.
A new standard for measuring the genuine usefulness of LLMs.
CONTENT PIPELINE: Latest content from AI Edge
New From AI Edge
YouTube: Claude 2.0: Everything That's Changed (FULL GUIDE). Watch now →
𝕏: How Elon Musk & SpaceXSI Use Grok Bot (copy this). Read
Skool: Free prompts, cheat sheets, PDF guides & more. Join here (100% free) →
See you next Wednesday,
The AI Edge Team
Want to sponsor or partner with us? Reach out at [email protected].