Sunday Rundown #130: Personal Agents & After-Sweepers

Sunday Bonus #90: Gemini Gem with access to the entire Why Try AI archive.

Happy Sunday, friends!

Welcome back to the weekly look at generative AI that covers the following:

• Sunday Rundown (free): this week’s AI news + a fun AI fail.

• Sunday Bonus (paid): an exclusive segment for my paid subscribers.

---

In case you missed it, here’s this week’s Thursday deep dive:

20 Free AI Image-To-Video Tools: Tested & Ranked [2026 Update]

If you’re consistently missing out on my emails, remember to check your “Promotions” tab and mark whytryai@substack.com as a “Safe Sender.”

---

I’ve been away last week, so we have two busy weeks’ worth of AI news to catch up on.

Let’s get to it.

👩‍💻 AI releases

1. ai.com launched a platform that lets you set up a personal AI agent to autonomously handle your tasks. (But read this before jumping in headfirst.)

2. Alibaba news:

   1. Qwen3.5 is an open-weight multimodal model that can handle complex tasks across text, images, and video, with support for 201 languages and dialects.

   2. Qwen-Image-2.0 is an image model that creates 2K-quality images, including posters, infographics, and photos with accurate text rendering.

3. Anam rolled out Cara-3, a real-time avatar model that turns any photo into an expressive, low-latency video. (Try the free demo.)

4. Anthropic news:

   1. Claude Code on Desktop can now preview your running apps, review your code with inline comments, and auto-fix errors before merging.

   2. Claude in PowerPoint is now also available on Pro plans and supports connectors to help you pull in context from third-party tools.

   3. Claude Sonnet 4.6 comes with upgrades to coding, agentic skills, reasoning, and design, as well as a 1M-token context window (in beta).

   4. Cowork launched on Windows with full feature parity, letting you automate multi-step desktop tasks with file access and plugin support.

5. ByteDance news:

   1. Seed 2.0 is a top-tier multimodal language model that handles text, images, video, and complex multi-step tasks.

   2. Seedance 2.0 is a best-in-class AI video model with native audio capable of complex choreography and multishot scenes.

6. Cursor launched a Plugin Marketplace with one-click installs for Notion, Figma, Neon, and other third-party apps.

7. Figma now lets you pull UI elements built with Claude Code into the Canvas and convert them into fully editable designs.

8. Google news:

   1. AI Overviews and AI Mode now provide on-hover link previews with more prominent icons, making it easier to jump straight to relevant websites.

   2. Gemini 3 Deep Think is a specialized reasoning mode for science, research, and engineering challenges. (Only for Gemini AI Ultra plans for now.)

   3. Gemini 3.1 Pro comes with doubled reasoning performance aimed at complex long-context tasks and practical applications.

   4. Lyria 3 lets you create 30-second music tracks from text or image prompts directly in the Gemini app, complete with lyrics and cover art.

   5. NotebookLM now lets you make revisions to its slide decks via simple prompts and export them to PowerPoint.

   6. Pomelli “Photoshoot” automatically turns product photos into studio-quality marketing assets using your brand’s colors and style.

9. KREA AI launched a Prompt-to-Workflow feature that auto-generates entire node setups from simple text instructions in Krea Nodes.

10. Manus now lets you access its agents via Telegram, so you can automate tasks on the go with full reasoning and tools.

11. Meta news:

    1. New “Animate” features let you animate your Facebook profile pictures, feeds, and more.

    2. Dear Algo lets Threads users tell the algorithm what content they want to see, adjusting their feeds accordingly for three days.

12. MiniMax open-sourced M2.5, a powerful coding and real-world productivity model that’s almost on par with top-tier competitors and significantly cheaper.

13. Moonshot AI launched Kimi Claw that lets you run OpenClaw AI agents in the cloud with access to 5,000+ community skills and 40GB storage.

14. NVIDIA now lets you set up OpenClaw AI agents for free locally on GeForce RTX and other consumer GPUs.

15. OpenAI news:

    1. Code Blocks in ChatGPT now let you preview and edit generated code more intuitively in split-screen or full-screen view without copy-pasting.

    2. GPT-5.3-Codex-Spark is an ultra-fast coding model running on Cerebras hardware at 1,000+ tokens per second. (Only for Pro plans for now.)

16. Replit launched Replit Animation that creates animated videos from text prompts without complex tooling or a learning curve.

17. Runway released an updated Story Panels workflow for building AI video sequences with consistent characters and scenes.

18. Tavus launched Phoenix-4, a real-time rendering model that creates emotionally aware AI avatars with realistic video and natural facial expressions.

19. WordPress introduced a built-in AI assistant that adjusts layouts, edits content, generates images, and translates text.

20. xAI is trialing Grok 4.2 in public beta with a rapid-learning architecture and weekly capability updates. (Select it manually in the dropdown.)

21. YouTube news:

    1. Conversational AI tool on smart TVs lets people tap “Ask” to get answers about the video without leaving the app. (In beta with a subset of users.)

    2. Personalized playlists let Premium users create an AI-curated playlist from an “idea, mood, or genre.”

22. z.ai released GLM-5, currently the best open-source reasoning model, almost on par with frontier models at a fraction of the cost. (Try it for free.)

---

🔬 AI research

1. Airbnb announced plans to add AI features for search, discovery, and customer support.

2. Apple’s iOS 26.4 adds CarPlay support for ChatGPT, Claude, and Gemini, so you can ask questions hands-free while driving. (In beta with a release in spring.)

3. Pika introduced AI Selves, an AI version of you that can remember things and act on your behalf. (Which is in no way creepy or dystopian, shut up.)

4. Reddit is testing AI-powered shopping carousels in search that pull recommendations from real user discussions.

5. T-Mobile announced live AI call translation in 50+ languages, built directly into the network with no need for special hardware. (In beta for eligible US subscribers.)

---

🔀 AI random

1. OpenAI started testing ads in ChatGPT for free-tier users in the US.

2. The Vatican introduced AI-assisted live translation for Holy Mass in 60 languages.

---

🤦‍♂️ AI fail of the week

Curling: You’re doing it wrong. (Somehow.)

---

💰 Sunday Bonus #90: Chat to my entire archive with a custom Gemini Gem

I’ve published over 300 posts on this newsletter.

That’s 3+ years of tool reviews, comparisons, guides, and opinions on just about everything AI-adjacent.

But good luck finding the exact answer you need right now.

My “Start Here” page is nice and all, but it can’t point you to my takes on a specific AI tool or give you tips on the fly.

Now that Google lets us hook up NotebookLM to Gemini, I built a custom Gemini Gem that has access to my entire, categorized post history.

You ask it a question, it answers based on what I’ve written and tested, and then links you to the most relevant posts: