🧠 Meta's Building a $200/Month AI Butler

Meta is prepping to launch Hatch, the consumer version of its internal OpenClaw agent, and internal documents show the company has floated charging up to $199.99 a month for a premium tier. That's not a typo. That's Claude Pro money, times two, for an agent trained to act inside DoorDash, Etsy, Reddit, Yelp, and Outlook.

The pitch is a dashboard where your agent shows off what it's built for you, like a fitness tracker or a trip itinerary it put together on its own. It's less "chatbot" and more "assistant with its own to-do list."

This is also Zuckerberg's clearest move yet to build a revenue stream that isn't advertising. If it works, expect every other platform to start asking what their agent is worth per month too.

Hatch is targeting a launch in the next few weeks, alongside a separate new model called Watermelon slated for October.

OpenAI's GPT-5.6 family, Sol, Terra, and Luna, is now live inside Kiro, AWS's spec-driven coding agent. The headline number: joint testing on Terminal-Bench 2.1 found GPT-5.6 Terra completing tasks in Kiro at roughly an 82% lower cost than before.

That's a vendor-run benchmark, so take the exact percentage with a grain of salt. But the direction is real, and it's part of a broader pattern this month of OpenAI slashing prices across the GPT-5.6 lineup, including an 80% cut to Luna and a 20% cut to Terra back in July.

For anyone building on top of these models, this is the kind of update that quietly changes your monthly bill more than any splashy feature launch would.

We ran an experiment pitting two popular Claude Code skills, Caveman and I-Have-ADHD, against each other on the same task: build a coffee shop website in one file. Both exist to cut down on the wall of text Claude tends to generate mid-task.

Caveman compresses Claude's language, stripping filler while keeping code and commands exact. It cut total output by 43% in our test. I-Have-ADHD takes a different approach entirely. It reorganizes what Claude says around your next action instead of just shrinking the words, and it cut output by 69%, skipping the mid-build narration completely and reporting once at the end.

The interesting twist: I-Have-ADHD isn't actually optimizing for brevity. In longer sessions, its rules deliberately restate progress so you don't lose your place, which can mean more words, not fewer. The real difference comes down to who's reading the output. If a machine or automation is consuming it, Caveman wins. If you're watching Claude work in real time, I-Have-ADHD is built for you.

A new tool from Vercel and Ora that scores how ready your website actually is for AI agents to use it, not just crawl it. It runs 100+ checks, shows you exactly where an agent gets stuck navigating your site, and gives one-click prompts to fix what's broken.

2. 💻 CanIRun.ai

This one solves the most annoying part of running AI locally: guessing whether your machine can actually handle it. Point it at your setup and it detects your GPU or Mac's unified memory, then grades over 100 open models, Qwen, Llama, DeepSeek, Gemma, and more, on what will run well, what's a tight fit, and what will just crash and burn.

You don't need to know how the model thinks. You need to know exactly what you want, the same way you'd need to be clear with a human collaborator. Nobody on your team can read your mind either. If you just say "make it good," the model has no idea what good means to you.

The fix is the same one that makes any collaborator better: show examples. Point to your own past work so the output stays on-brand, or share references you like for inspiration. Clarity beats cleverness every time.

Last Friday's session had a great turnout, and we're opening the doors to everyone again this week. Join us live, build in real time, and get your questions answered on the spot.

👉 Save your spot today!

Can’t make it live? Don’t worry, we’ll send out the recording after the session to everyone who RSVP’ed.

Event details:

📅When: Friday at 10AM PT/1PM ET
📍Where: Virtual, Google Meet

After you RSVP, you’ll be sent a link with the event details. Hope to see you there!