Hey Sonny here! ππΌ
I'm going to let you in on something β this newsletter was not written the way you think it was. Keep reading.
π₯ The Big One
This newsletter was written by AI agents
No really. The edition you're reading right now was researched, written, and QA'd by an AI agent pipeline. Let me show you exactly how it works.
I built a system called OpenClaw β an open-source multi-agent framework that runs on my Mac mini 24/7. For this newsletter, three specialized agents work together:
-
Scout (research agent) scours the week's AI coding news, finds 8-10 stories with official images and sources
-
Ink (writing agent) takes Scout's research and writes the full edition using our template, in my voice
-
Pulse (QA agent) runs a pre-send checklist β catches factual errors, broken links, and formatting issues before anyone sees it
The draft gets pushed straight into our Mission Control dashboard where my team and I review and edit. Then we copy the final version to Kajabi and hit send.
Total human time? About 20 minutes of review β the agents handle the heavy lifting.
Here's the thing most people get wrong about AI agents: they try to build one super-agent that does everything. That's like hiring one person to research, write, and proofread. It doesn't work.
Each of my agents has specialized skills and runs on the right model for their job. Scout needs broad research capabilities. Ink needs deep writing craft. Pulse needs obsessive attention to detail. Specialization beats generalization β for humans and AI agents.
"The breakthrough wasn't building one smart AI. It was building a team of specialists that hand off work to each other β just like a real content team." β Sonny
And here's the other thing we shipped this week that makes this whole pipeline seamless:
Real-time collaborative editing in Mission Control
We built Google Docs-style collaboration directly into our YouTube script editor and newsletter editor.
-
Live cursors and presence powered by Liveblocks β see exactly where your teammate is editing in real-time
-
Document persistence and sync via @convex-dev/prosemirror-sync on top of our TipTap editor
-
Scripts and newsletters auto-sync every keystroke β no save button, no stale data, no "which version is latest?"
-
Multiple team members can edit the same document simultaneously
-
Agents can push content directly via API β Scout researches, Ink writes, and it appears in the editor instantly
The real unlock? When the newsletter pipeline finishes writing, the draft shows up in Mission Control in real-time. My team and I can jump in and edit simultaneously β with live cursors showing exactly where the other person is working. No copy-pasting. No version conflicts. Just open the doc and start reviewing.
"We went from filesystem storage and manual copy-paste to real-time sync where humans and AI agents all write into the same document." β Sonny
β‘ What shipped this week
1. One engineer rebuilt Next.js on Vite β in a week, for $1,100
A single Cloudflare engineer used Claude to rebuild 94% of the Next.js API from scratch. It's called vinext β Vite-based, deploys to Cloudflare Workers. 4x faster builds. 57% smaller bundles.
This isn't a toy. This is a production-ready Next.js alternative. One person. One week. About $1,100 in AI tokens. We're in the "AI lets one person build what used to take a team" era.
"One engineer rebuilt 94% of the Next.js API surface in approximately one week using Claude as a coding assistant." β Cloudflare Blog
Read the full Cloudflare blog post β
2. GitHub Copilot CLI is now GA
Terminal-native AI coding β and it's not just autocomplete anymore. This is a full agentic coding assistant running in your terminal.
Supports GPT-5.3-Codex, Claude, and Gemini. Multi-file editing, project-wide context, all from the command line. Every paid Copilot subscriber gets this now.
"Terminal-native coding agent, now generally available for all paid GitHub Copilot subscribers." β GitHub Blog
Check out GitHub Copilot CLI β
3. Claude Code Agent Teams β multi-agent coding is here
Anthropic shipped Agent Teams for Claude Code. A lead agent assigns tasks, teammate agents claim work with file-locking to prevent conflicts, and they code in parallel across your codebase.
This is what I've been building with OpenClaw β but now it's built right into Claude Code. The pattern is the same: specialization beats one agent trying to do everything.
"Lead agent assigns tasks, teammates claim work with file-locking. Parallel coding across your codebase." β Cobus Greyling
Read about Claude Code Agent Teams β
4. Next.js 16 ships a built-in MCP server
Your Next.js app now has a /_next/mcp endpoint that lets AI agents see your running app's errors, routes, and component state.
Your AI coding assistant can now look at your actual running application while it helps you code. No more copy-pasting error messages. The agent sees what's broken in real-time.
"The built-in MCP endpoint lets AI agents inspect your running app's errors, routes, and component state." β Next.js Docs
Read the Next.js MCP guide β
5. Vercel Queues enters public beta
Vercel finally has native background jobs. Vercel Queues is a durable event streaming system built on their Fluid compute infrastructure.
No more bolting on Upstash, Redis, or BullMQ for background processing. If you build order flows, email workflows, or AI pipelines on Vercel β this is worth evaluating now.
"Durable event streaming for background jobs, natively on Vercel." β Vercel Community
π§° Worth your time
-
AI Coding Agent Landscape β Excellent breakdown of editor assistants vs. repository agents. Claude Code for hard problems, Copilot for daily work. Worth reading if you're trying to decide which tools to invest in.
-
Notion 3.3 Custom Agents + Workers β Notion shipped autonomous AI workers that run on schedules and triggers, built on Vercel. If you're building tools for teams, this is a new distribution channel worth watching.
-
How to Build Your Own AI Agent Team (OpenClaw Tutorial) β My full walkthrough on setting up OpenClaw on a VPS, connecting tools via MCP, and running specialized agents 24/7. This is exactly how I built the newsletter pipeline above.
π Want to build your own AI agent team?
Everything I showed you today β the newsletter pipeline, the multi-agent system, the real-time editing β runs on OpenClaw. It's open-source, runs locally or on a VPS, and connects to all your tools via MCP.
I put together a full tutorial showing how to set it up from scratch:
βΆοΈ Watch: How to Build Your Own AI Agent Team β
My weekly message to YOU
Before I sign off β a personal note.
Due to the ongoing conflict and regional security developments, things have been tense here in Dubai. We've experienced emergency alerts, temporary airspace closures, and shelter advisories from authorities. Our team is safe, but we're prioritizing staying indoors with our families right now.
Because of this, there will be a short delay in our content timeline on YouTube and Instagram. We truly appreciate your understanding and patience during this time, and we'll keep you updated as things evolve.
Stay safe out there, wherever you are.
What's the one task in your workflow that you'd automate first?
Hit reply β I want to hear it. I read every single one.
Talk soon PAPAFAM,
Sonny ππΌ
ππ½ Don't forget to follow me across socials!
Responses