This week felt like a turning point. Not a "cool new feature" kind of week — a the entire game just changed kind of week.
Let me break it all down.
🔥 The Big One
Cursor Just Launched Fully Autonomous Cloud Agents
This is the one that stopped me mid-scroll.
Cursor just dropped Cloud Agents — fully autonomous AI coding agents that run on cloud VMs. Not autocomplete. Not "helpful suggestions." Actual agents that code for 25-52+ hours, self-test their work, record video demos of the result, and deliver finished PRs.
Read that again.
Here's the part that should wake you up: 35% of Cursor's own PRs are now AI-generated. They're not just selling the tool — they're building their company with it.
Oh, and they just hit a $29.3 billion valuation.
The AI coding war isn't coming. It's here. And the companies winning it are the ones actually using what they build. — CNBC
The agents spin up cloud VMs, install dependencies, run the full dev loop, test everything, and hand you a PR. You review. You merge. That's it.
This is what agentic development looks like. Not a copilot. A coworker.
🛠️ What I built this week
Every week I build with AI agents. Not just about them — with them. Here's what shipped from my desk this week:
-
Mission Control dashboard — Built a full internal ops dashboard from scratch. Kanban board with drag-and-drop for reminders, a rich text newsletter editor with real-time sync (ProseMirror + Convex), and a script editor for YouTube videos. The best part? My AI agents push content directly into it via API. No more context-switching between twelve tabs.
-
KaneAI demo store — Spun up a Next.js 16 e-commerce app for an upcoming sponsored video. Full checkout flow — browse → product → cart → checkout → confirmation. Designed specifically for AI testing demos. Built the entire thing in one session with AI pair programming.
-
Agent cost optimizations — Running 8 AI agents 24/7 gets expensive if you're not smart about it. Routed routine crons to cheaper models (Haiku and Codex instead of Opus), extended cache retention to 1 hour, reduced heartbeat frequency. Cut costs significantly without losing a single capability. Work smarter, not harder.
-
Newsletter API with body push — This newsletter you're reading? An AI agent pushed the draft — fully formatted with images, blockquotes, and highlights — directly into my Mission Control editor in one API call. No more manual copy-paste. The agents write, the editor catches, I review and ship.
I'm not building AI tools. I'm building with AI tools. There's a massive difference — and it's where the real leverage is. — Sonny
⚡ What shipped this week
1. GPT-5.3-Codex GA + The Codex App
OpenAI went all-in on agentic coding. The Codex App is basically a command center for AI agents — parallel workstreams, agent supervision, clean diffs, and a new Codex-Spark mode for real-time coding. Think of it as mission control for your AI dev team.
2. Claude Code Security
Anthropic just made Claude into a security engineer. It scans your entire codebase for vulnerabilities, then patches them automatically. In preview alone, it found 500+ vulnerabilities. Cybersecurity stocks dropped 6-9% the day this launched.
When an AI can find and fix security vulns faster than your security team... that's not a tool. That's a shift. — The Hacker News
3. GitHub Copilot Multi-Model (Agent HQ)
Copilot now supports Claude, Codex, and Gemini with an Auto mode that picks the best model per task. No more vendor lock-in. Best tool for the job, every time.
4. Xcode 26.3 — Agentic Coding Comes to Apple
Claude and Codex are now native in Xcode. That means every major IDE now has agentic AI built in. VS Code, JetBrains, Cursor, and now Xcode. There's nowhere left to hide from this shift.
5. WebSockets in OpenAI Responses API
Persistent connections for agent runs. 20-40% faster execution. If you're building long-running agents, this is a big deal — less overhead, faster feedback loops.
🧰 Worth your time
-
OpenAI's guide to long-running agents — Shell tool + Skills + Compaction patterns. If you're building agents that run for more than a few minutes, this is required reading.
-
"Something Big Is Happening" essay — 80M+ views. Broke dev Twitter. My take? The devs who feel threatened aren't using the tools yet. Once you actually build with AI agents, the fear disappears and the leverage becomes obvious.
-
How to Setup OpenClaw Securely That Runs 24/7 - The SAFE way! — My full tutorial on deploying OpenClaw on a VPS the right way. Secure, always-on, isolated from your personal machine. This is the video people keep DMing me about.
🚀 Want to set up your own OpenClaw?
If you want to set up your own multi-agent system but do it securely — running 24/7 on a VPS instead of your laptop — I made a full tutorial walking through the exact setup. From zero to 8 AI agents running around the clock, the right way!
▶️ How to Setup OpenClaw Securely That Runs 24/7 - The SAFE way! →
My weekly message to YOU!
If you're not experimenting with agentic workflows yet — this is your week to start. Pick one task. Hand it to an agent. Watch what happens.
The compound effect is unreal.
What's your current AI coding setup — Cursor, Copilot, Claude Code, something custom? Hit reply, I want to know what's actually working for you.
I read every single one.
Talk soon PAPAFAM,
Sonny 👋🏼
👇🏽Don’t forget to follow me across socials!
Responses