Happy Sunday, !
Welcome back to your weekly AI news roundup.
In case you missed it, here’s this week’s Thursday post:
Note: If you’re consistently missing out on my emails, remember to check your “Promotions” tab and mark whytryai@substack.com as a “Safe Sender.”
Here’s what happened in AI last week:
👩💻 AI releases
Adobe news:
Creative Agent can now handle multi-step edits across the company’s flagship cloud and desktop apps like Illustrator and Premiere.
Firefly became a full AI creative studio that lets you use Adobe’s own tools along with partner models like GPT Image 2.6 and Kling 4.0 on one canvas.
Anthropic news:
Claude Duet, Anthropic’s proactive assistant that works on tasks in the background, is now open to all paid users on the US waitlist. (I’m not jealous.)
Claude Fable 5.2 is even better at coding, reasoning, and vision tasks and, at the same time, cheaper and faster than its older cousin.
Claude Meetings has Claude join your team’s video calls so it can draft documents, whiteboards, artifacts, and more based on your conversation.
Claude Sheets and Claude Whiteboards let paid users build shared spreadsheets and whiteboards and export them to tools like Excel or Miro.
Interactive connectors for apps like Canva and Notion let them open right inside Claude so it can make edits in real time.
Automattic launched Spacefast, a free workspace for your AI assistant to publish the apps and websites it builds with share links.
Edge0 open-sourced Edge0-145B, a coding and reasoning model that brings “frontier AI capabilities” to your laptop, running offline in a free desktop app.
Google news:
Gemini 4 Carbon, Google’s strongest model, is rolling out to AI Ultra subscribers in the Gemini app and also powers their AI Mode in Search.
Gemini 4 Hydrogen is a fast model built for everyday coding and knowledge work tasks and is especially good at sticking to instructions.
Gemini Intelligence is coming to more Android phones and can build custom widgets and troubleshoot your phone.
Google Wallet now connects to the Gemini app, so you can ask Gemini to chart your spending or spot overlooked subscriptions. (US only for now.)
Meta released Muse Ember, its smallest and cheapest multimodal model built for high-volume jobs like coding subtasks and document analysis.
MiniMax open-sourced H3-Image, a 4K image model that can keep track of up to 16 reference images and finish a picture in about ten seconds.
Nuance Labs launched Emma, an AI video avatar that reacts with facial expressions and acts like a person on a call. (Try for free.)
OpenAI news:
Businesses in ChatGPT lets your US business claim its profile and design branded agents that answer customer questions inside ChatGPT.
ChatGPT Wallet allows ChatGPT to buy things for you without seeing your card details, and it includes a Face ID or fingerprint check before approval.
🔬 AI research
ElevenLabs is rolling out early access to ElevenAI, a free desktop app that captures dictation and meeting notes using context from your past conversations.
Google opened a waitlist for Tabletop, a Labs experiment that creates shared group spaces for things like expense splitting, trip planning, and more.
📖 AI resources
“Building always-on agents” [GUIDE]: the building blocks behind agents that wake up to handle scheduled tasks and background jobs (by Anthropic).
“Fable 5.2: The Self-Improvement Loop Keeps Loopin’” [VIDEO]: a deeper look at the current frontier models like Fable 5.2 and Gemini 4 Carbon by AI Explained.
“MotionGraphicsBench” [BENCHMARK]: a set of 100 tasks by Figma and Stanford that tests how well AI models can animate a static graphic.
“Pressure.dev” [TOOL]: start a debate with multiple models to pressure-test your ideas.
[sponsored]
🤦♂️ AI fail of the week
Plenty of fails here, but the synchronized group high-fiving of the air tops them all.
🔐 All my paid goodies in one place
Take a peek at what’s behind the paywall:
