Good Morning! Here's what I have for you in today's newsletter:
Google DeepMind introduces Genie 4, a model with sound and live controls
OpenAI brings iCloud Mail and Calendar plugins to every ChatGPT plan
CopilotKit releases OpenDuet, an open-source proactive self-hosted agent
The Claude Code video editing guide for busy creators
4 new AI tools worth trying today
AI MODELS
Google DeepMind introduced Genie 4, a world model that creates interactive scenes you can explore in real time. Access starts with trusted testers, followed by Google AI Ultra subscribers in Project Genie and early-access API partners.
Change the scenery, add objects, or speak with characters as you explore. Gemini plans multi-step events to help the world stay consistent through those changes.
It’s Google’s first real-time world model with generated audio, including ambient sounds, music, background noise, and dialogue alongside the visuals.
Built on Gemini Omni, it simulates collisions, aerodynamics, and human movement. Google also plans to use these environments to test agents and robots before they operate in the real world.
A generated world gets more interesting when you can change it while standing inside it. Genie 4 adds sound and live interaction, with potential uses beyond entertainment, including simulated tests for agents and robots. Access remains limited, so how well those worlds hold together outside Google’s demonstrations is still worth watching.
TOGETHER WITH FRONT
A fast AI answer isn't always a coordinated one. When AI resolves B2B issues without looping in the teams responsible for them, confidence outpaces alignment. A new Harvard Business Review Analytic Services briefing paper, sponsored by Front, looks at closing that gap.
AI PLUGIN
OpenAI added iCloud Mail and Calendar plugins to ChatGPT, allowing it to search Apple email and manage calendar events. Both are available on every ChatGPT plan across web, desktop, and iOS.
The Mail plugin can find emails, summarize your inbox, manage messages and drafts, send replies, and look up iCloud contacts.
The Calendar plugin reads, creates, and updates events in iCloud Calendar. Changes appear in your iPhone’s Calendar app and other apps synced to that account.
Connect each plugin separately using an Apple Account app-specific password. Supported addresses include @icloud.com, @me.com, and @mac.com. Gmail and Outlook accounts in Apple Mail need separate connections.
The booking details are in an email, but your availability is in the calendar. Connecting both lets ChatGPT work through that familiar back-and-forth in one conversation. The distinction is the account, though: using Apple Mail doesn’t make a Gmail inbox an iCloud inbox, so check which service holds your messages.
AI AGENTS
CopilotKit released OpenDuet, an open-source agent that flags what needs attention and works between check-ins. You host it, choose the model, and connect agents that support the AG-UI protocol.
Morning briefs separate items that need you from those that can wait. It can also create and update designs, dashboards, and web pages.
Give it a separate computer and browser, and tasks can continue after your laptop shuts down. Sending, paying, signing, and deleting require approval.
Access is available through web, mobile, Slack, and Microsoft Teams. OpenDuet is MIT-licensed and in early development, with a free CopilotKit Intelligence project handling threads and memory.
OpenDuet offers a self-hosted take on agents that check in first and keep working between conversations. Teams can adapt the code and choose their model, but they also take on the setup and hosting. It’s early software, and its use of CopilotKit Intelligence for memory and threads is part of that setup to understand.
HOW TO AI
Claude Code coordinates Whisper for transcription and ffmpeg for cutting, captions, and export, and you review the edit and save your preferences for next time. You’ll need Claude Code, an Apple silicon Mac, and a recording of yourself speaking.
Step 1: Install the tools
Check claude --version and ffmpeg -version. Install ffmpeg with brew install ffmpeg if needed. Create a Python environment with python3 -m venv ~/.video-edit, then run ~/.video-edit/bin/pip install mlx-whisper. Test the import using that environment’s Python. The first transcription downloads a speech model.
Step 2: Set up the edit folder
Record one or two minutes on a subject you know. If you stumble, repeat the full sentence so the retake is easy to spot. Create ~/my-video/output and copy your recording into ~/my-video. Start Claude there and ask it to write CLAUDE.md with your caption style, colors, and pacing. Require cut-list approval, protect the original recording, and keep generated files in output/. Save reusable preferences separately from one-off notes.
Step 3: Review the cuts
Ask Claude for word timestamps, silence detection, and a proposed list of pauses and retakes to remove. Read each row and name anything you want kept, for example “Keep row 3 because it sets up the joke.” After approval, have it transcribe the edit again. Listen around the joins yourself to catch clipped words or abrupt pauses.
Step 4: Add captions and a hook
Request a 1080x1920 version with framing that keeps your face or screen content visible. Add short captions timed to your voice and an opening title that disappears by five seconds. Keep text clear of platform controls. Give specific notes and save the preferences you want reused.
Step 5: Export and check
Export H.264 at 1080x1920 and 30 fps, with AAC audio and loudness targeting -14 LUFS. Ask Claude to compare the final transcript with the approved wording and report the file settings. Watch the export on your phone, then save the tested process as a video-edit skill. For the next video, copy your preferences and the skill folder into a new folder.
P.S. You can access all my AI courses, trainings, prompts library, and $500+ worth of AI tool discounts here.
Google added live audio recording to Gemini Notebook’s mobile app, with sessions up to one hour long. Record a lecture or meeting, then use it as a source alongside other material, including photos of handwritten notes.
Cognition introduced FrontierCode 2.0, a coding benchmark built around one question: should this pull request be merged? Tasks are larger and deliberately leave engineering decisions to the model. The highest score so far is 45%.
Parallel launched its Decisions API for answering decision questions using current web data. Requests cost $0.001 each, with 5,000 free to start. Listed uses include checking facts, updating stale records, and classifying leads.
🎨 Cursor: Added Annotate to design mode for editing elements, drawing directly on pages, and checking layouts at different screen sizes.
🖥️ Manus: Websites built with Manus can now be installed as desktop apps through Chrome’s install button.
🧊 SpAItial: Released Echo-3 to turn phone captures into metric-scale 3D environments, covering rooms, multi-floor buildings, and outdoor spaces.
🤖 Reason: Added messaging between agents, including Grok Bot, Muse, Claude Code, and Codex, so they can contact your teammates’ agents too. A free trial is available.
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
✓ $500+ worth of AI tools