Sunday Rundown #142: AI Podcasts & Mirror Jester
Vote on tools you'd like me to cover during upcoming Live & Learn sessions.
Happy Sunday, friends!
Welcome back to the weekly AI news roundup.
In case you missed it, here’s this week’s Thursday deep dive:
If you’re consistently missing out on my emails, remember to check your “Promotions” tab and mark whytryai@substack.com as a “Safe Sender.”
🗞️ AI news
Here’s what happened in AI this week:
👩💻 AI releases
Alibaba released Qwen3.7-Max, its new flagship agent model built for coding, workflow automation, and long-horizon tasks. (Try for free at Qwen Chat.)
Amazon launched Alexa Podcasts, so you can request a custom audio episode on any topic that Alexa researches, writes, and narrates. (Alexa+ users only.)
Cohere open-sourced Command A+, a mixture-of-experts agentic reasoning model that can run on just two H100 GPUs. (Try for free on Hugging Face.)
Cursor launched Composer 2.5, a coding agent that matches frontier model performance at a fraction of the cost.
Google news (mainly from I/O 2026):
AI Mode was upgraded with Gemini 3.5 Flash and now includes search agents and can build mini-apps to handle tougher questions and tasks.
Antigravity 2.0 is a new standalone desktop app that can orchestrate many AI coding agents at the same time and schedule background tasks.
Gemini 3.5 Flash is the company’s best-in-class model for coding and long-horizon agent tasks that’s also four times faster than frontier alternatives.
Gemini Omni Flash can create and edit videos from any combination of audio, images, text, and video inputs.
Gemini Spark is a 24/7 personal AI agent that handles multi-step tasks across your Calendar, Docs, Drive, Gmail, Sheets, and more.
Pomelli gained agentic features that can help businesses create brand identities, brand books, and full websites in just a few clicks.
Project Genie can now use Google Street View to let you explore an AI-generated version of any real-world location. (Gemini Ultra users for now.)
Stitch Agent lets you build, stream, and steer app designs in real time from text, voice, code, or existing design files. (Try for free.)
SynthID expanded its AI watermarking across Chrome, Gemini, Google Search, and Pixel to help people spot when content was made with AI.
Granola launched Briefs that generate a pre-meeting summary of what was discussed last time and what to focus on just before you join.
Ideogram launched an MCP server to let AI chatbots and agents generate and edit images directly in the chat.
Krea released LoRA fine-tuning for Krea 2 in beta, so you can train the model on your own images to lock in a specific style, character, or object.
Leonardo AI launched 3D Generation, which can turn a prompt or image into an exportable .glb model without leaving the platform.
Microsoft launched an improved Bing Image Search that organizes image results into AI-labeled groups that help you find what you’re after faster.
OpenAI news:
Appshots lets Mac users send any open app window straight into Codex without copying anything manually.
ChatGPT for PowerPoint lets you generate and edit Microsoft PowerPoint slides using natural language. (Try for free during global beta)
Verify is a new tool that lets you check whether an image was made or edited by OpenAI’s models. (Try it for free.)
Stability AI open-sourced the Stable Audio 3.0 family of music models that can generate 6-minute tracks and were trained on fully licensed data. (Try it for free.)
🔬 AI research
Apple previewed AI-powered accessibility updates like Vision Pro eye-tracking wheelchair controls, natural language Voice Control, and more.
Figma announced its Design Agent that works inside your file, follows your design system, and makes edits directly on the canvas. (Request early access.)
Google news:
AI Studio is adding a mobile app so you can start AI projects on your phone and continue on your desktop. (Pre-register for the app now.)
Universal Cart will let you add items via Gmail, Gemini, Google Search, and YouTube into an AI cart with automatic price alerts and stock notifications.
Microsoft previewed MagenticLite, a lightweight multi-agent framework optimized to run browser and file tasks with small models.
Spotify news:
Fan-made covers will let Premium subscribers create AI-generated covers and remixes of songs from participating Universal Music artists.
Personal Podcasts can create AI audio episodes on any topic from a prompt, PDF, or link on a recurring schedule. (Premium subscribers only.)
📖 AI resources
“Two Rival Bets on AGI: Google I/O Highlights” [VIDEO]: fascinating look at the different pathways to AGI pursued by different labs, from AI Explained.
🔀 AI random
Anthropic hired Andrej Karpathy (ex-Tesla and OpenAI co-founder) to work on Claude pre-training and speeding up frontier model research.
OpenAI had its reasoning model autonomously disprove an 80-year-old “unit distance” geometry problem, with results verified by Fields medalist Tim Gower.
🤦♂️ AI fail of the week
I wanted distorted, surrealist version of a person reflected in the mirror, following their motions. This…is not that.
📹 Live & Learn: What shall I tackle next?
After a brief break, I’m planning to return to my weekly Live & Learn sessions again.
Let me know what types of tools you’d like me to test:
Thanks for reading!
If you enjoy my work, here’s how you can help:
❤️Like this post if it resonates with you.
🔄Share it to help others discover this newsletter.
🗣️Comment below, —I love hearing your opinions.
Why Try AI is a passion project, and I’m grateful to those who help keep it going. If you’d like to support me and unlock cool perks, consider a paid subscription: