Good morning! Here's what's happening in AI today:
Alibaba launches Qwen3.8-Max, a new bar for coding and cowork
xAI adds full video analysis capability to Grok
ChatGPT Voice goes fully hands-free on macOS and Windows
How to build an AI second brain that gets smarter every time you use it
4 new AI tools worth trying today
AI MODEL
Alibaba released Qwen3.8-Max, a new model the company positions as a fresh bar for coding and collaborative work, with a demo showing it designing chips nonstop for 12 hours straight.
The model is built specifically around coding tasks and cowork scenarios, rather than general-purpose chat, positioning it against dedicated coding models instead of broad assistants.
Alibaba's own demo runs the model on a genuinely long, unattended task, hours of continuous chip design work rather than a quick scripted example.
It slots into the growing field of large coding-focused models competing directly with the coding-specific releases coming out of US labs.
A model that can run for 12 hours on one task without falling apart is a different claim than fast single-turn code completion, it's a bet on long-horizon reliability. If that holds up outside a controlled demo, it matters for anyone running agentic coding workflows that need to keep working unattended. The real test is whether it holds that reliability on messier, real-world codebases rather than a curated showcase task.
AI TOOL
xAI rolled out the ability for Grok to analyze any video, a capability confirmed directly by Elon Musk, extending the assistant's reach beyond text and images into full video comprehension.
Grok can now process video content directly, opening it up to tasks like summarizing footage, answering questions about what's happening on screen, or pulling specific details out of a clip without a human watching it first.
The feature works across the kinds of video people already share and post daily, not a separate upload tool bolted onto the side.
This puts Grok in a smaller group of assistants that handle video natively, rather than requiring a transcript or a manual description first.
Video is the one format most AI assistants still handle poorly, usually reduced to a transcript before anything useful happens. If Grok can actually watch and understand a clip, that opens real use cases: content moderation, research, sports and news analysis, all without someone sitting through the footage first. Whether the analysis holds up on longer or lower-quality video is the real test still ahead.
AI VOICE
OpenAI brought ChatGPT Voice to macOS and Windows, letting people work entirely hands-free on desktop, a capability the company demonstrated directly rather than just describing in a release note.
Voice now runs natively on desktop instead of being confined to the mobile app, so a full conversation with ChatGPT no longer requires typing at all.
The demo shows it working while someone's hands stay occupied with something else entirely, which is the actual point of a hands-free mode rather than a novelty.
It's available now on both major desktop operating systems, not staged as a Mac-first or Windows-first rollout.
Voice interfaces only matter if they replace typing in situations where typing is inconvenient, cooking, driving, working with your hands full, and a desktop-native version means that's no longer limited to a phone screen. This closes a real gap between ChatGPT's mobile and desktop experience. Whether people actually adopt hands-free workflows at a desk, where typing is usually still faster, is the open question.
HOW TO AI
Your best ideas are scattered right now. Notes app, browser tabs, an old Claude chat you closed and will never find again. Every session starts cold, you rebuild context from memory, then explain yourself from zero again.
The fix isn't a better prompt. It's giving Claude a place to actually live.
The stack: two free tools, one job each.
Obsidian stores everything as plain text files on your own machine, notes link to each other and form a visible graph of how your thinking connects. Claude sits on top of it, reading the vault, filing new information, answering questions across everything you've ever saved.
Download Claude Desktop from claude.com/download and Obsidian from obsidian.md, and create a vault.
Step 1: Open the door
Inside Obsidian, turn on Community plugins, install Local REST API, copy the API key it generates. Then connect Claude to it:
claude mcp add-json obsidian-vault '{ "type": "stdio", "command": "uvx", "args": ["mcp-obsidian"], "env": { "OBSIDIAN_API_KEY": "PASTE-YOUR-KEY-HERE", "OBSIDIAN_HOST": "127.0.0.1", "OBSIDIAN_PORT": "27124" } }'
Step 2: Load yourself into the brain
A connected vault that knows nothing about you is a filing cabinet, not a brain. Have Claude interview you instead of typing your whole life story by hand:
You are setting up my second brain. Interview me ONE question at a time to
build my profile. Ask about: who I am and what I do, my goals for this year,
how I want you to talk to me, my strengths and weaknesses, and my current
projects. Wait for each answer before the next question. When finished, write
everything into a file called CLAUDE.md at the vault root, structured with
headers, so you load it automatically every session.Answer like you're briefing a co-founder, not filling out a form.
From there, the rest of the build is project folders (Inputs, Process, Outputs, Feedback), a rulings file that turns corrections into permanent rules instead of repeated ones, and a daily scheduled task that files everything overnight while you sleep. Same subscription, completely different machine. Read the full guide.
P.S. You can access all my prompts, workflows and AI trainings if you upgrade here
Vercel launched Passport for general availability to Enterprise customers, giving teams native IdP support, verified identities across apps, and the ability to monitor access and create bypasses when needed.
Firecrawl open-sourced pdf-inspector, the engine behind its /parse endpoint, classifying any PDF and extracting clean markdown locally in roughly 0.002 seconds per page without waiting on OCR.
MiniMax shipped MiniMax-H3, now publicly available for anyone to run, joining the wave of open-weight releases that no longer require an invite or a waitlist to actually test.
💾 LM Studio: Run DeepSeek V4 Flash 0731 locally on a 128GB Mac, or through the cloud, at a fraction of the usual footprint for a model this size.
🎬 Remotion: HTML in Canvas now ships with 10 new filters and 6 dynamic transitions for building video directly in code.
🐟 Sakana AI: The updated Namazu large language model is now available as a direct API.
🏷️ Morphic: Tag clips and mark favorites, then ask Copilot to pull up exactly what you need from the library instead of scrolling.
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