Good morning,
What happens when you point a camera at a video and watch it transform into a world you can walk through? That's no longer science fiction. Fei-Fei Li's World Labs just made it possible with their Marble AI - and it works on your Quest 3 today.
Meanwhile, Cursor's autonomous agents just built a full web browser from scratch. It wrote over a million lines of Rust code in about a week.
Topics of the day:
AI transforms any video into explorable 3D worlds
Cursor agents build a browser with 1M+ lines of code
Why US companies are quietly using Chinese AI models
How 1 in 4 workers now use AI on the job
The Shortlist: Apple revamps Siri with on-device AI, UCLA's new Alzheimer's detection model, Sequoia doubles down on Anthropic
I also added a new section where I share what I'm reading and using right now, including tools, articles, and people worth your attention.
World Labs Turns Any Video Into Explorable 3D Worlds
What's happening: Fei-Fei Li's World Labs launched Marble, an AI that transforms 2D images and videos into persistent, interactive 3D environments you can explore in VR.
In practice:
Upload a photo, video, or text prompt and Marble generates a downloadable 3D world - not just a flythrough, but an environment you can edit, modify, and export.
Works with Vision Pro and Quest 3 today - every generated world is VR-ready out of the box.
Outputs industry-standard formats (Gaussian splats, meshes, videos) that drop straight into existing 3D workflows.
Robotics teams can now generate unlimited training environments instead of building physical test spaces.
Bottom line: This is spatial computing's "ChatGPT moment." World Labs raised $230M from a16z, NEA, and investors including Eric Schmidt at a $1B+ valuation - and Marble shows why.
AI Agents built a full web browser
What's happening: Cursor's autonomous coding agents built FastRender, a complete web browser with custom rendering engine, writing over 1 million lines of Rust code across 1,000 files in roughly one week.
In practice:
Hierarchical agent system: planners create tasks, workers execute code, judges verify completion - like a self-organizing engineering team.
Agents had access to WhatWG and CSS-WG specification files as reference material, letting them implement web standards from scratch.
The browser actually works - renders Google.com and complex sites with "legible and mostly correct" output, though with visible glitches.
This mirrors Claude Code's sub-agent architecture, suggesting the pattern is becoming standard for complex AI coding tasks.
Bottom line: We're past the "AI writes functions" phase. Autonomous agents can now coordinate to build entire applications - complete with custom rendering engines that took human teams years to develop. When AI can build browsers, what can't it build?
A Startup built on 150,000 lines of AI-generated code
What's happening: Startup BoothIQ published a deep-dive on building a company with 150,000 lines of AI-generated code, sharing a transparent look at the good, the bad, and the ugly of extreme automation.
In practice:
AI can dramatically speed up development for specific tasks like frontend changes or generating code in concise, well-defined languages.
Your role shifts from line-by-line coding to being the architect, as AI struggles with high-level structure and needs human guidance to avoid creating a mess.
Expect to manually debug complex issues, as current models fail to grasp concepts like concurrent processes and will get stuck chasing ghosts without your help.
Bottom line: Going all-in on AI-generated code offers a massive productivity boost, but it isn't autonomous. Your most important job becomes setting the architectural direction and maintaining the quality bar.
Why US companies are quietly using Chinese AI models
What's happening: US tech companies like Pinterest and Airbnb are quietly adopting Chinese open-source AI, citing lower costs and better performance than closed-off US alternatives.
In practice:
You can slash AI operational costs by evaluating open-source models for tasks like customer service, which can be up to 90% cheaper than proprietary APIs.
Fine-tuning your own models on company data gives you a performance edge - Pinterest saw a 30% accuracy boost in its recommendations by customizing open-source AI.
Instead of defaulting to big-name US models, explore platforms like Hugging Face to find specialized open models that better fit your specific business problem.
Bottom line: The best AI model for your business isn't always the one with the biggest marketing budget. Open-source alternatives offer a powerful, low-cost way to build a competitive advantage.
1 in 4 workers now use AI on the job
What's happening: Roughly one in four American workers now use AI frequently on the job, a new poll finds. The data shows AI is moving beyond tech roles and into everyday tasks across finance, education, and even retail.
In practice:
Use it to refine your communications, like cleaning up a quick email to a client or drafting a solid first pass of a recommendation letter.
Bridge knowledge gaps in real-time by using an AI assistant to get quick answers on topics outside your core expertise when a customer or colleague is waiting.
Accelerate data analysis by having AI summarize dense reports or synthesize large documents to pull out key insights in minutes, not hours.
Bottom line: AI adoption is past the tipping point for the American workforce. Professionals who treat it like a standard productivity tool will save time and create more leverage in their roles.
What I read/use at the moment
Tools, articles, and people worth your attention.
CallMe - A Claude Code plugin that calls your phone when tasks finish or need decisions. Solves the "waiting at your computer" problem when using Claude Code - kick off a long task and get a call when it's done or blocked. Surprisingly useful for longer builds.
Huxe - AI-powered audio briefings that connect to your calendar and email. I’ve been using this app daily since i installed it. It works kinda like your personal newspaper x EA/PA, but in Podcast format. It looks at your day, connects the dots, does some research, and shares news around subjects you choose. What makes it stand out is that you can interrupt and ask follow-up questions mid-playback, turning passive listening into an interactive conversation.
Clawd - Peter Steinberger's personal AI assistant running on a Mac Studio with 25+ tools (email, WhatsApp, smart home, etc).
The Shortlist
Figure showed its 02 humanoid robot stacking and sorting dishes with high precision, demonstrating how its general-purpose AI model can learn new, complex physical tasks from training data alone.
NewsGuard found that major chatbots generate misinformation 35% of the time, a sharp increase from 18% last year, signaling growing reliability risks for businesses using them for content and research.
Orchard Robotics raised $22M to expand its vision AI technology for agriculture, which helps farmers more precisely monitor crop health and improve yields.
This newsletter is where I (Kwadwo) share products, articles, and links that I find useful and interesting, mostly around AI. I focus on tools and solutions that bring real value to people in everyday jobs, not just tech insiders.
Please share any feedback you have either in an answer or through the poll below 🙏🏽
