Good morning! Here's what's happening in AI today:
Google Flow Music just got a real upgrade with Lyria 3.5
HeyGen turns any doc or idea into a two-host video podcast
Tavus lets anyone build a persistent AI assistant with no code
How to turn any idea into a working app with Stitch and Claude Code
4 new AI tools worth trying today
AI MUSIC
Google DeepMind launched Lyria 3.5, the new music model now powering Flow Music, bringing more expressive vocals, richer arrangements, and BPM control with exportable stems for full-length songs.
Vocals get real dynamic range now, not the flat, robotic singing older music models tend to produce.
Arrangements are built to flow naturally across a full track instead of looping the same few bars, a common giveaway of AI-generated music.
BPM control plus exportable stems means the output isn't just a finished clip, it's something a producer could actually pull apart and remix.
Most AI music tools stop at "impressive for a demo." Stems and BPM control push this into something closer to a real production tool, the kind of output someone could drop into an actual project instead of just sharing as a novelty. That's a different bar than most music models have cleared so far.
AI TOOL
HeyGen launched Video Podcast, turning any doc, link, or idea into a two-host video show complete with studio scenes, multi-cam cuts, and B-roll, ready in minutes.
Feed it a source, a doc, a link, a rough idea, and it builds an actual two-person conversation around it, not just narrated audio.
Studio scenes and multi-cam cuts are handled automatically, the kind of production work that normally means an editor and a few hours.
This goes further than audio-only podcast tools, the pitch is specifically "this is a show you can publish," not just a script read aloud.
Podcast production has always meant either learning to edit or paying someone who already knows how. This compresses that whole process into minutes, which matters most for anyone who has ideas worth turning into content but no time or budget for a production team.
AI AGENT
Tavus introduced PAL Maker, a no-code way to build a PAL, a persistent AI application that sees, hears, acts, remembers, and responds to one person or a thousand, built entirely through conversation.
No code required. You describe the PAL you want, and it gets built from that description alone.
Persistence is the actual feature here, this isn't a one-off chat, it remembers you across sessions the way a coworker or tutor would.
The examples given range from a tutor that remembers your progress to something that checks in on a family member, a genuinely wide range of use cases from one underlying tool.
Most AI assistants forget you the moment the session ends. A PAL is built specifically not to. That distinction matters most for the use cases where continuity is the entire point, checking in on someone, tracking progress over weeks, not just answering a single question well.
HOW TO AI
A designer builds something in Figma. A developer opens it, eyeballs the spacing, guesses the hex codes, rebuilds it by hand. Screen one looks right. Screen ten has a padding value that's off, and nobody notices until a user does.
Google Stitch and Claude Code now talk to each other directly. Describe the app, Stitch designs it, Claude Code builds it, and one file, design.md, keeps both in sync the whole way through.
Step 1: Describe What You Want Type a description at stitch.withgoogle.com, Stitch generates a full interface, layout, typography, spacing, and components, consistent because one model designed all of it in one pass.
Step 2: Export design.md Every export includes a plain-English file describing the color palette, typography, and spacing scale, not code, not JSON, just readable sentences.
Step 3: Connect the Two Tools Install Stitch's MCP server so Claude Code can call it directly, no switching between a browser tab and a terminal.
Step 4: Build From the Same Source Point CLAUDE.md at design.md, and every component Claude Code writes after that loads the same design system automatically.
Three Ways to Actually Use This
Description to Working App
Describe the app, let Stitch design it, hand the export straight to Claude Code, idea to working code in under 30 minutes.
Talk Through a Change Instead of Typing It
Voice Canvas lets you describe an edit out loud, Stitch adjusts in real time or asks a clarifying question instead of guessing.
Skip the Design Step Entirely
A free collection of 70-plus pre-made design.md files, extracted from Stripe, Linear, Notion, and others, means you can skip generation and just build straight from an existing design system. Access the full guide with tons of other prompts and use cases here.
P.S. You can access all my prompts, workflows and AI trainings if you upgrade here
Perplexity open-sourced Numbat, an agent-detection and response layer that works across different agent harnesses, giving security teams visibility into agent activity with the ability to block actions before they execute.
Notion shipped high contrast mode, built for anyone who finds Notion's muted default design hard to read, sharpening text, icons, and borders.
Onton announced Ontology 1, a new search model claiming at least 2.7x more accuracy than the best ecommerce search engines, learning on its own with no retraining or fine-tuning required.
🌐 TinyFish in Grok Build: Gives Grok's coding agent real search, fetch, and browser access to the live web.
🖼️ Reve Chrome Extension: Save any image from the web straight into your Albums or References.
🧹 Ideogram Object Remover: Erases any object, text, or distraction with zero visible trace.
🏗️ Rox Apps: Build a custom app on your existing customer data in minutes.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial