Happy Sunday, !
Welcome back to your weekly AI news roundup.
In case you missed it, here’s this week’s Thursday post:
Stuck on a Task? Get Unstuck with AI.
TL;DR If you’ve been putting off a task, ask AI to help spot the few blocking decisions only you can make, so you can finally get moving.
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
Alibaba released Qwen4-Max, the “most capable model in the Qwen family to date,” which is especially good at coding and visual reasoning.1
Anthropic merged Claude Design into the main Claude app, so new designs will now be made as Claude artifacts. (The standalone site closes December 14.)
Audible launched audio discovery, short AI-narrated previews offering a quick take on what listeners liked about a specific audiobook.
Google news:
AI Mode in Search now has temporary threads for questions you don’t want saved to your history or used to personalize results.
Gemini 4 Neon is a fast frontier model for coding and analysis on par with Opus 5.5 and GPT-6.2 Sol, available to paid subscribers in the Gemini app.
Gemini for Mac can now take on repetitive tasks by clicking and typing through your apps in its own window while you keep working.
Gemini Notebook now turns your notes and sources into interactive sites on demand, and you can also share the results with others.
Genie 4 lets you generate interactive worlds and reshape them mid-session with a prompt. (Exploring ChatGPT and I tried something like this last week.)
Tasks in Gemini is expanding for free to six countries, calling businesses and booking appointments on your behalf.
HeyGen launched Avatar 6, which can turn a single photo into a realistic talking video avatar without length limits.
Microsoft news:
MAI Code 2 is a coding model that nearly matches Claude Opus 5.5 on everyday coding tasks for a fraction of the cost.
MAI Image 2.7 is a frontier-level model that can generate and edit images inside the new Canvas workspace.
Office in Copilot brings Excel, PowerPoint, and Word together in the Copilot app and lets you work on files together with teammates.
OpenAI news:
ChatGPT dictation now turns speech into text in real time as you talk, rather than in one big chunk after you finish.
GPT-6.2 Astra and GPT-6.2 Sol are OpenAI’s newest model upgrades, with Astra built for the hardest professional projects and Sol for everyday work.
Perplexity news:
Brain can now create and update personal skills based on your past sessions, so Perplexity Computer learns how you prefer your tasks done.
Perplexity Computer got a so-called “Multiagent effort level,” where agents that run different frontier models split up tasks and check each other’s work.
Proton released Lumo Desktop, a privacy-first AI assistant for your computer that lets you decide per task what it can do without permission.
Replit launched Missions, which lets you set a budget and goal to have an AI agent swarm build and iterate toward it automatically.
Slack is becoming an automation hub that lets users automate tasks with Slackbot and third-party AI agents via messages, schedules, and more.
Suno launched Visuals, which turns your songs into either lyric videos or short clips that lip-sync to the words.
🔬 AI research
Google opened a waitlist for Stitch Loop, an always-on agent that keeps independently working on your product to fix and improve it. (Join the waitlist.)
Microsoft news:
Autopilot is a cloud-based always-on work agent that proactively manages tasks and coordinates work across your chats, emails, and meetings.
Copilot Code will let you build apps and interactive dashboards and then refine them using natural language.
Today is an upcoming command center in the Copilot app and Outlook that flags what needs your attention and can prepare send-ready drafts.
📖 AI resources
“Research acceleration: The jagged frontier” [REPORT]: a look at how much of OpenAI’s own research work its AI agents can now handle on their own.
“World 2030” [SIMULATION]: an interactive scenario that shows alternative ways the US economy, jobs, and prices could develop up to 2030.
🔀 AI random
Anthropic officially became the first purely AI company to go public, trading on Nasdaq under the “ANT” ticker.
🤦♂️ AI fail of the week
The girl is, like, way too chill about her mom being the Dark Wizard.
Want to follow my ongoing “AI Shorts experiment”? Check out this dashboard.
🔐 All my paid goodies in one place
Not a paid subscriber yet? Take a peek at what’s behind the paywall:
For what it’s worth, the model is currently #3 on my FunBench benchmark.



