Presented by
Good morning, AI enthusiasts. The agentic web just got its biggest platform yet. Google turned Chrome into an AI assistant that can navigate websites, fill forms, and take actions on your behalf starting with tasks most people do dozens of times a week.
The shift from "AI that answers questions" to "AI that does things" is accelerating faster than expected, and the browser wars just became the agent wars.
In today's recap:
Chrome's AI assistant handles tasks automatically
Moltbot runs without Mac minis via Cloudflare
Create AI-generated promo videos with Remotion
AI startup raises $180M for human-like learning
4 new AI tools, prompts, and more
Google Chrome gets an AI brain upgrade
Screenshot
Recaply: Google just rolled out major Gemini updates to Chrome on Windows, macOS and Chromebook Plus, adding a persistent side panel, native image editing, deep Google app integrations, and agentic auto browse that handles multi-step tasks on your behalf.
Key details:
The new side panel keeps Gemini accessible without switching tabs, allowing users to multitask while the assistant handles separate workflows like drafting emails, comparing information across sites, or generating images with Nano Banana integration.
Connected Apps tie Gmail, Calendar, YouTube, Google Shopping, and Google Flights directly into Chrome, with AI Pro and Ultra subscribers in the US getting auto browse that can scroll, click, fill forms, check bills, manage subscriptions, and even add items to shopping carts while staying within budget.
Auto browse requires user confirmation for sensitive actions like purchases or social media posts, with Google Password Manager handling credentials when authorized, and the system pausing to let users take over tasks at any point.
Personal Intelligence arrives in coming months with opt-in app connections and conversation memory for context-aware assistance, while Google's Universal Commerce Protocol support ensures AI agents can execute transactions on behalf of users.
Why it matters: Chrome isn't just adding AI features. It's shifting from a passive tool you use to an active agent that does things for you. While competitors offer chatbots in sidebars, Google's betting that agentic browsing with Connected Apps and auto browse will make Chrome indispensable for complex workflows.
PRESENTED BY HUBSPOT
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
CLOUDFLARE
Run viral AI agent Moltbot without buying a Mac mini
Cloudflare
Recaply: Cloudflare just launched Moltworker, an open-source project allowing developers to run Moltbot, the viral self-hosted AI agent formerly known as Clawdbot, directly on its cloud platform without dedicated hardware like Mac minis.
Key details:
Moltworker combines Cloudflare's Sandbox SDK, Browser Rendering, R2 storage, and AI Gateway to provide complete infrastructure for running the assistant remotely, with the $5 monthly Workers paid plan as the minimum requirement alongside free-tier options for other services.
The tool emerged as Cloudflare's stock jumped from $183 to $210 following viral enthusiasm around autonomous AI agents, though the company notes this is a proof of concept showcasing platform capabilities rather than an official product.
Security concerns dominated Hacker News discussion, with developers warning that giving agents access to email and filesystems creates massive prompt injection attack surfaces, and the project attracts technically inexperienced users who may not understand the risks.
Moltworker uses AI Gateway for centralized model management and Zero Trust Access for authentication, while Browser Rendering handles web automation through a CDP proxy, avoiding the need to run Chromium in containers.
Why it matters: The viral Clawdbot wave exposed a gap between hype and reality. People want AI assistants that actually do things, but setting up self-hosted agents on dedicated hardware isn't realistic for most users. Cloudflare's jumping on that gap with cloud infrastructure that makes it easier, though the security issues don't disappear just because it's running on their servers instead of your Mac mini.
TUTORIAL
Create AI-generated promo videos with Remotion
Recaply: In this tutorial, you will learn how to generate professional promotional videos for your app or website using Claude Code's Remotion agent skills, eliminating hours of manual video editing by describing your vision in natural language.
Step-by-step:
Install Claude Code via command line by running “curl -fsSL https://claude.ai/install.sh | bash” on Mac or “irm https://claude.ai/install.ps1 | iex” on Windows PowerShell, then create a dedicated project folder and navigate to it with cd Documents/your-folder-name.
Add the Remotion agent skills to Claude Code by running “npx skills add remotion-dev/skills --no-ui”, which gives Claude the ability to understand and execute video editing tasks autonomously.
Initialize your Remotion project by running bun create video my-app and following the on-screen setup instructions, then activate the Remotion skill by typing "activate remotion skill" in the terminal.
Go to the my-app/public folder, delete any existing files, and upload your brand assets like logos or custom images that you want Claude to use in the video generation.
Prompt Claude with your video requirements by describing the brand name, copy text, and desired style, and Claude will autonomously build a multi-layered video with scene breakdowns, animations, and transitions that you can preview in Remotion Studio at localhost:3001.
Pro tip: After Claude generates the initial video, you can iterate by asking for specific changes in the terminal or manually adjust the code in Remotion Studio for fine-grained control over timing and effects.
AI RESEARCH
AI startup raises $180M to make models learn like humans
Flapping Airplanes
Recaply: San Francisco startup Flapping Airplanes just secured $180 million at a $1.5 billion valuation from Google Ventures, Sequoia, and Index to build AI that learns like humans without ingesting half the internet, focusing on data efficiency over compute scale.
Key details:
The founding team estimates humans are 100,000x to 1,000,000x more sample efficient than current models, since humans exceed AI capabilities despite seeing only a few billion tokens by adulthood compared to models trained on essentially all accessible history.
Andrej Karpathy endorsed the launch on X, noting the conventional wisdom that new research startups can't compete was wrong when OpenAI started against Google, wrong again when new labs challenged OpenAI, and remains wrong today with breakthroughs still possible despite massive scaling progress.
Flapping Airplanes is pursuing paradigm-shifting research on fundamentals like loss functions and gradient descent itself, with the team including world-class researchers and rising stars who are IMO, IOI, and IPhO medalists, though they're not currently trying to commercialize.
Sequoia partner David Cahn distinguished this as one of the first labs taking a research-first approach over scaling, arguing we're 2-3 breakthroughs away from AGI intelligence rather than needing to dedicate maximum societal resources toward scaling up today's LLMs.
Why it matters: The AI industry has been locked in a compute arms race, with every major lab convinced that bigger clusters and more data are the only path forward. Flapping Airplanes is betting the opposite, that fundamental research breakthroughs matter more than server buildouts. The $180M at $1.5B valuation for a lab with no product and no commercialization plans shows how seriously investors are taking this contrarian bet. If they're right, AI progress could accelerate dramatically without requiring exponentially more compute.
NEWS
What Matters in AI Right Now?
Perplexity signed a $750 million three-year cloud deal with Microsoft to access Foundry models from OpenAI, Anthropic, and xAI via Azure.
Google DeepMind rolled out Project Genie to Google AI Ultra subscribers in the U.S., an experimental prototype powered by Genie 3 that generates photorealistic, interactive 3D environments from text prompts in real time.
Tencent open-sourced Hunyuan 3.0-Instruct, a text-to-image model with support for multi-turn conversations and controllable image generation.
Allen AI introduced Theorizer, a multi-LLM framework that automatically reads scientific literature and synthesizes structured theories with laws, scope, and evidence extracted from thousands of papers.
Contextual AI announced Agent Composer, adding orchestration capabilities with multi-step reasoning and multi-tool coordination to their enterprise context layer platform.
Google DeepMind published AlphaGenome in Nature, an AI system for predicting genetic variant effects with applications in genetics research and disease understanding.
TOOLS
Trending AI Tools
💻 Moltbook - AI agent social network for sharing and discussing content
🦞 OpenClaw: Viral AI agent assistant that works 24/7, formerly known as Clawdbot
🎮 Project Genie: Google DeepMind's experimental prototype that generates photorealistic, interactive 3D environments from text prompts
🌍 ChatGPT Translate - OpenAI's translation tool supporting 40+ languages
PROMPTS
Generate Production-Ready n8n Workflows
Adopt the role of an expert n8n Workflow Architect, a former enterprise integration specialist who spent 5 years debugging failed automation projects at Fortune 500 companies before discovering that 90% of workflow failures come from unclear requirements and missing context. You developed an obsessive attention to detail after a vaguely defined automation requirement cost a client $2M in lost revenue, and now you can translate any automation idea into production-ready n8n workflows with surgical precision.PS: This is not the full prompt. Click the button below to access the complete prompt.
Have a favorite prompt? Tell us about it or rate today’s prompt, click here.
EVENTS
Raycast Stockholm Hackathon: Feb 06, 2026 • Stockholm, Sweden
Build India: Feb 15, 2026 • Bengaluru, India
Claude Code Meetup Danver: Feb 17, 2026 • Boulder, Colorado
MCP Connect Day: Feb 05,2026 • Paris
Paris AI Hackathon: Feb 07-08, 2026 • Paris
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
