Good morning! Here's what's happening in AI today:
Grok now builds and publishes full apps from a single prompt
OpenAI shipped two transcription models built for real-world noise
Perplexity's Personal Computer just landed on Windows
How to give Claude the ability to watch any video on multiple platforms
4 new AI tools worth trying today
AI TOOL
Grok launched Build Mode, letting you describe an app and get back something live with its own domain, no hosting setup, no export step, no second tool required.
The whole loop happens in one place: you describe it, Grok generates it, it's live. That's the entire pitch, and for once it's not overselling.
Right now it's locked to SuperGrok Heavy subscribers. Everyone else is watching from outside.
This puts Grok in the same lane as Replit and Vercel's v0. xAI charging a premium tier for it tells you they think builders will pay to skip the boring parts.
Getting an idea live used to mean weeks: hosting, deployment, a domain to register. Build skips straight to the end of that list, if you're paying for SuperGrok Heavy. Everyone else is watching from outside for now, worth checking back when that gate opens up.
AI TOOL
OpenAI introduced two new transcription models, GPT-Live-Transcribe for the moment audio happens and GPT-Transcribe for the pile of recordings already sitting in your drive.
GPT-Live-Transcribe targets real-time use, live captions, meeting notes, anything where a five-second lag ruins it.
GPT-Transcribe is the batch version, tuned for accuracy over speed on audio that's already finished.
Both are built to handle the audio that actually breaks transcription tools: accents, background noise, clipped phrases, jargon nobody trained the model on.
Transcription has quietly been the weak link under a lot of AI tools you already use. Garbled names. Missed context. Output you end up fixing by hand anyway. If these models hold up on messy, real audio instead of studio-clean clips, that's the difference between a transcript you trust and one you re-check line by line.
AI AGENT
Perplexity launched Personal Computer for Windows, a local agent harness that works across your files, your connected apps, and the open web from one system instead of five open tabs.
Research, code, browse, build, all inside the same workflow. No copy-pasting between four different tools to finish one task.
Because it runs locally, it reaches your actual files directly. Not just what a browser can see.
Paired with Model Council (see Essential Bytes), the direction is obvious. Perplexity isn't trying to be a better search engine anymore. It's trying to be where your work happens.
Most AI tools live inside a browser tab. They can't touch anything on your actual machine. This one can, at least on Windows. If your day already bounces between local files, live research, and code, that's fewer handoffs, not a new set of steps to learn. Different product than search with a chat window bolted on.
HOW TO AI
Claude can't watch video. Anthropic doesn't have a video model yet. Most workarounds just pull the transcript and hand it over, which misses half of what's actually happening, the graphs, the on-screen text, the moment someone holds up a prop. A video is pictures and words. Transcript-only tools give Claude one of them.
There's a free skill that gives Claude both.
Step 1: Install the Skill
One command in Claude Code, two minutes total.
/plugin marketplace add bradautomates/claude-video
/plugin install watch@claude-video
Step 2: Paste Any Video Link
/watch https://youtu.be/[video-id]
Claude downloads the video, pulls screenshots and the transcript, then reads both together, not just the words, the actual frames too.
Step 3: Claude Pulls Both Halves Automatically
Screenshots and transcript, processed together, no manual work on your end.
Step 4: See What It Actually Catches
Ask for structured notes on a longer video, and it comes back referencing things a transcript alone would never catch, exact numbers read off a chart, a joke revealed by an on-screen title card, specs shown but never spoken aloud.
Three Ways to Actually Use This
1. Break Down Why a Hook Worked:
Point it at a video doing numbers in your niche and ask why, it goes second by second through the first 5 seconds.
2. Turn a Lecture Into Structured Notes:
Point it at any talk or workshop and ask for notes instead of a transcript dump, references specific charts and slides at exact timestamps.
3. Reverse-Engineer a Competitor's Ad:
Point it at a competitor's demo, it breaks down the opening claim, the proof, the offer, and catches the on-screen visuals a transcript would miss.
Three of the six use cases are here. The full guide covers the last three, plus a second pipeline that takes everything Claude picked up from watching and turns it into a brand new video.
P.S. You can access all my prompts, workflows and AI trainings if you upgrade here
Fish Audio launched S2.1 Pro publicly. Clones a voice from 5 seconds of audio, runs twice as fast as Cartesia, and costs a sixth of what ElevenLabs charges.
ElevenLabs expanded ElevenAgents across more channels. Agents now answer over SMS and Telegram, and handle tickets through Intercom and Freshdesk.
OpenAI quietly open-sourced the Codex Security CLI, letting anyone scan repositories, track findings across runs, and add security checks straight into CI/CD.
🎬 Agent Opus: Turns your coding agent's output into a publish-ready video with music, voiceover, and captions.
🪞 Mirage Avatar X: AI avatar model built around identity preservation, works in vertical and horizontal video.
⚡ Liquid AI LFM2.5 Encoders: Bidirectional encoders that stay fast at long context, even on CPU.
🔀 Vercel Sandbox: Now supports forking with Sandbox.fork(), inheriting config and env vars from the source.
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