Good Morning! Here's what I have for you in today's newsletter:
Google brings Lyria 3.5 to the Gemini App, its best-sounding music model yet
ComfyUI adds MiniMax H3 Max, post-trained for speed and precision
Cursor lets cloud agents run on your own auto-scaling infrastructure
Bring your Claude Code project to Codex without losing months of setup
4 new AI tools worth trying today
AI AGENTS
Google brought Lyria 3.5 to the Gemini App, calling it its best-sounding music generation model yet, with more expressive vocals, richer musical arrangements, and new templates to start a track from.
Templates give users a concrete starting point for a track, ready to shape into their own idea instead of typing a description from a completely blank page.
Users can choose between short or longer tracks, and select or describe a genre before deciding between vocal or instrumental output.
More expressive vocals and richer arrangements are the two specific improvements Google points to over its previous music model.
Choosing vocal or instrumental output, then setting genre and track length before generating, gives real creative control over the final result. Templates add a second entry point into the tool, letting someone build from an existing starting shape rather than only from a text description.
AI MODELS
ComfyUI brought MiniMax H3 Max live, a model post-trained by Fal specifically for speed and precision, now generating directly inside ComfyUI's node-based workflow with no separate platform required.
The model is designed for consistent, repeatable output rather than broader creative range, aimed at production use over experimentation.
It generates natively as ComfyUI nodes, integrated directly into the platform's existing pipeline structure.
Native node generation supports direct chaining with upscaling and compositing operations.
Post-training aimed specifically at speed and precision positions this model for production pipelines, where consistent, repeatable output matters most. It runs directly inside ComfyUI's existing node system, integrated into the platform's production pipeline.
AI TOOLS
Cursor added the ability to run cloud agents on your own infrastructure, including pools of machines that automatically scale with demand while the agent loop stays inside Cursor.
Agents can now reach internal services or specialized hardware directly, opening up systems that live inside a team's own environment.
Machine pools scale automatically based on demand, adjusting capacity in real time as agent workload changes.
The agent loop itself continues running inside Cursor even though the actual compute happens on the user's own infrastructure.
Running agents on internal infrastructure gives a team's own private systems and specific hardware direct reach into the agent's work. Auto-scaling machine pools handle capacity automatically, adjusting on their own as agent workload shifts throughout the day.
HOW TO AI
Months of teaching a tool how you work is why people stay put. Codex can find an existing Claude Code setup and bring most of it across in minutes.
Step 1: Find your existing setup
Go to Settings, then Import, then Claude Code. It scans your machine, tick what you want brought over. Nothing gets moved or deleted. CLAUDE.md renames to AGENTS.md automatically.
Step 2: Bring over the leftovers
Hooks and output styles have no Codex equivalent, and custom subagents often need a second pass.
npx -y skills add openai/skills --skill migrate-to-codex --agent claude-codeRun the migrate-to-codex skill against my Claude setup, plan and dry-run first before applying anything
Step 3: Read your own rules back
Open the folder in Codex and paste:
Read my instructions file and tell me in plain English what it asks you to do. Rewrite it native to Codex, keeping every rule. Find anything that only works in Claude Code and give me the Codex equivalent, or tell me there isn't one.Step 4: Run both tools together
/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/codex:setupStep 5: Let them catch each other's mistakes
Build this, then ask Codex directly what's wrong with the result before telling me it's done
P.S. You can access all the AI trainings (including the full version of this one), prompts and workflows if you upgrade.
Notion Calendar introduced smart scheduling, recommending times that work for everyone when adding people to a new event, so conflicts can be previewed and a slot picked without back-and-forth.
Google rolled out flexible usage limits in Gemini Notebook, adding deferred artifact generation and shifting limit resets to every 5 hours instead of daily.
Browser Use introduced native support for WebMCP, giving agents direct HTTP calls instead of relying on a UI built for humans, making interaction fully deterministic.
💻 GitHub Copilot: this week's release adds GPT-6 Astra, Fable 5.1, and Gemini 3.8 Flash as model options, plus multi-root workspaces and Agent Merge in public preview.
💬 Slack: Add to Slack launches with 10 platforms, including OpenAI, Lovable, and n8n, letting anyone deploy working AI agents into their channels in a few clicks.
🌐 Airtop: automate your work with just words, AI-driven cloud browsers that log into any site and navigate the web freely.
💰 HubSpot: Breeze AI Invoice Prioritization is now generally available, ranking open invoices by revenue and customer risk and drafting collection emails automatically.
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