🧠 Finally a Robot that Folds Your Laundry

Moonshot released Kimi K3: 2.8 trillion parameters, reads up to a million tokens at once, understands text and images natively. In blind tests, developers preferred it over top US models for coding, including some of Anthropic's and OpenAI's own.

Why you should care: open weights land by July 27, 2026. Moonshot is upfront that K3 still trails Claude Fable 5 and GPT 5.6 Sol overall. But a frontier-adjacent model you can host yourself changes the math for anyone building on top of APIs.

You can try it now at Kimi.com or via platform.kimi.ai.

The Codex Micro is a keypad for people running OpenAI's coding AI, Codex. Six lights show you what each of your AI agents is doing, a dial adjusts how hard the AI thinks, a joystick fires shortcuts like "review this code."

If you juggle multiple AI agents, losing track of them is a real problem. This is OpenAI testing whether people want a physical remote for that. Its currently out of stock, but if you’re interested you can buy it here.

Sunday's robot, Memo, folded laundry across 31 homes it had never seen, hitting 99.1% success. Human judges rated the folds 4.72 out of 5. The wild part: it can learn a brand new fold from just one example and repeat it anywhere.

Why it matters: most robot demos are cherry-picked. Sunday published the actual conditions and success rate, not just a highlight reel. That's the bar to hold every AI demo to.

Think of this as a shortcut to running powerful open AI models (like the Kimi model above) without needing your own expensive computer setup. You just call an API and it handles everything behind the scenes. It also works as a drop-in replacement if you're already using OpenAI's tools, so switching over is easy. Good if you want to experiment with different AI models without committing to one.

A tool that lets your AI agents actually read live websites instead of relying on outdated information. Give it any web link and it hands back clean, usable text, or you can tell it exactly what data to pull out, like prices or company details. Useful if you're building anything that needs current, accurate information from the internet.

Take a look here.

Fortune sat down with a few company executives and asked a simple question: why do so many AI projects work great in testing and then fall apart when the whole company starts using them?

The answers had almost nothing to do with the AI itself. One exec said the real trick is letting teams try lots of small experiments, but being strict about which ones actually get scaled up, since not every good experiment deserves a full rollout. Another pointed out that teams often get excited about shipping a cool AI feature and forget to ask if it's actually solving a real business problem.

But here’s what we think: A lot of companies don't even have their own processes written down clearly. So when they add AI on top of a messy, undocumented process, they expect magic, and instead get nothing. That applies just as much if you're running a one-person business. If you can't explain your own process step by step, an AI can't run it for you either.

Read the full article here.

This is your final chance to RSVP for the Live Q&A that Kushank will be hosting this morning. All details below:

🗓️ When: Friday, July 17th @ 10AM PT (1PM ET)
 📹️ Where: Virtual

RSVP takes ten seconds, and the question you send sets the agenda. So bring it on!