🧠 How To Turn Your Living Room Into a Video Game

ChatGPT Work is a new agent built into ChatGPT that can take action across your apps and files, not just answer questions about them. It's powered by Codex and the newly released GPT-5.6, and it's built to stay on a project for hours, breaking a big goal into smaller steps and working through them without you babysitting every move.

Connect it to Slack, Gmail, Google Drive, or your CRM and it can pull context from all of them to build a deck, a doc, or a full report. You can even hand it recurring work through Scheduled Tasks, like turning new Slack messages into an updated status doc every Monday morning.

It's rolling out today for Pro, Enterprise, and Edu users, with Plus and Business getting it over the next few days. The desktop app now folds Chat, Work, and Codex into one place, complete with a built-in browser so it can pull in live web context too.

Anthropic just launched Reflect, a dashboard that tracks how you've been using Claude and turns it into something you can actually look back on. It shows when you use Claude most, what kind of work you tend to bring to it, and lets you set quiet hours or a nudge to take a break after a stretch of use.

The interesting part is the framework underneath it. Anthropic is scoring your usage against four dimensions they call the "4D AI Fluency Framework," things like whether you're settling your own strategy before delegating, or reworking drafts in your own voice instead of shipping them as-is.

It's in beta now for Free, Pro, and Max users who have Memory turned on, and it deliberately skips incognito chats and anything tied to a health integration.

1X unveiled a 25-degree-of-freedom, tendon-driven hand for its NEO robot, and the pitch is that it finally removes the hardware ceiling that's been holding robotics back. These hands are force-transparent, meaning every joint reports back what it's touching instead of going numb the moment it grips something.

In practice that means NEO can do things a basic robot gripper never could: assemble Lego, sort grapes by color, use a screwdriver, pour tea without spilling it, and wash its own hands since the whole thing is IP68 waterproof and food-safe.

1X says they've already got hundreds of these hands off the production line and are aiming for 10,000 built this year. That's the number that actually matters, since a great hand nobody can manufacture at scale doesn't move the industry forward.

Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.

Creator Karen Cheng wired an actual rotary phone from the 1920s to an AI agent that replies to you through a mechanical display. Basically a dumbphone that talks back, minus the doomscrolling.

She built the whole thing using Cursor and walked through it publicly, so if you've ever wanted a genuinely offline-feeling way to interact with AI, this is a fun blueprint to steal from.

Here’s a quick look at how this works!

Kushank spent some hours building a fun little game. Using Claude Fable 5, his laptop's camera, an HDMI cable, and a TV, he was able to put together a game called Cardio Surfer.

It plays just like Subway Surfers, except instead of swiping, you jump to jump, duck to duck, and step side to side to switch lanes. Your webcam tracks your movement and turns it into the game. If you'd rather stay seated, it also works with a phone or keyboard.

Here’s a quick demo of how fun this actually is!

Designer Emil Kowalski combed through Apple's WWDC videos and distilled what he found into 17 design and motion principles, packaged as a free skill you can point any coding agent at. Use it to review work you've already shipped or to keep something you're currently building from looking generic.

Quick pro tip before you install anything off GitHub: most "skills" today are just a folder of instructions your agent reads before it starts working. You don't need to touch the code.

Just run the install command in your terminal from inside your project folder, and the agent picks up the rules automatically from there.

Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.