This past week was absolutely wild in AI. Like, genuinely hard to keep up. And one creator just dropped a weekly roundup that makes it all digestible.
Matt Wolfe from FutureTools put together a packed breakdown of everything that shipped this week, and honestly, some of these updates are worth paying attention to.
The big headline? The expert highlights that Anthropic has now shipped 74 releases in just 52 days. The site Product Compass even made a calendar showing the pace, and it’s relentless. But the twist here isn’t just volume. It’s what they shipped.
🔥 The highlights from Anthropic’s shipping spree:
- Computer Use is now live. Claude can literally control your mouse and keyboard. You give it a task, and it clicks around your desktop to get it done. It’s slow right now (like 5 minutes for a 10-second task), but combined with the Dispatch feature on mobile, you can text your computer from anywhere and have it work on your behalf.
- Co-work Projects lets you organize work inside Claude with custom instructions and file context. Think of it like ChatGPT’s projects feature, but inside Claude’s desktop app.
- Auto Mode for Claude Code finally stops the annoying permission prompts. No more walking away, coming back, and finding Claude waiting to ask “can I run this terminal command?” Harmless commands just run now.
💡 Pro tip from the video: To use Computer Use, you need a paid Claude plan ($20/month minimum), and you have to manually toggle it on in Settings > General > Desktop App.
Google wasn’t sitting still either. Gemini 3.1 Flash Live rolled out across the API, enterprise, search, and the Gemini app. The creator showed it recognizing him through a webcam, reading his OBS setup via screen share, and walking him through software in real time. He also demoed Google’s experimental browser that generates entire web pages on the fly using Gemini 3.1 Flash. Type “Taco Cat Parade” and it builds a full page in seconds. No memory or saving yet, but the speed is impressive.
🎵 Music AI got a big week too:
- Google’s Lyria 3 Pro now generates tracks up to 3 minutes with intros, verses, and choruses
- Suno 5.5 lets you clone your own voice and generate songs with it
- Mistral’s Voxtral TTS dropped as an open-weights text-to-speech model you can run locally, and it holds up against 11 Labs in blind tests
⚡ Rapid fire from the roundup:
- Sora is dead. OpenAI killed the app, the API, and the video generator entirely. Disney exited their deal the same day, reportedly with zero advance notice.
- OpenAI also shelved their planned “adult mode” for ChatGPT.
- Early ChatGPT advertisers say they can’t prove the ads actually work. Meanwhile, OpenAI launched richer product discovery in ChatGPT, which feels like the groundwork for paid product placements.
- A leaked Anthropic blog post revealed Claude Mythos, a new model tier above Opus with dramatically higher coding and cybersecurity scores. Anthropic themselves called it “kind of scary” from a security perspective. The post was pulled, so take it with a grain of salt.
- Wikipedia banned AI-generated articles but still allows AI for copy editing and translation.
- The Figure 03 robot walked into the White House. Nobody was harmed.
This is one of those weeks where every major player shipped something meaningful. Check out the full video for demos of Computer Use, the Google real-time browser, Suno voice cloning, and more.