Meta just told employees its Hatch agent "can do anything you can do online" — which means the battle for your digital life just became a three-way fight between your phone, your browser, and your AI.
The Summary
- Meta is testing "Project Hatch," an AI agent that can book restaurants, order food, fill out forms, buy things, and access your Instagram, positioning it as "personal superintelligence for everyone"
- Hatch gets its own computer to execute tasks autonomously, competing directly with OpenAI's OpenClaw, Google's Gemini Spark, and Anthropic's computer-control agents
- Meta is framing Hatch as a revenue opportunity, not just an R&D flex — this is about monetizing their AI stack beyond ads
The Signal
Meta's internal memo describes Hatch as an agent "always working on your behalf to help achieve your goals and improve your life, your health, your relationships, your personal finances and more." Strip away the corporate therapy-speak and what you have is this: Meta wants to be the operating system for your digital errands.
The timing matters. OpenAI's OpenClaw went viral earlier this year by showing what autonomous agents could actually do when you gave them computer access. Google launched Gemini Spark. Anthropic shipped computer-control features. A startup called Instinct is getting buzz for letting users call or text their AI assistant. The agent race is no longer about who can answer questions fastest. It's about who can *do things* for you while you're doing something else.
"Hatch has its own computer so it can do anything you can do online, like filling out forms, buying something, or performing deep research."
Here's what separates this from Meta AI, the chatbot they've been shoving into every app: Hatch isn't reactive. It doesn't wait for you to ask. The memo says it "can remember more than most AIs," which in practice means it's building a persistent model of your preferences, habits, and goals. It books the restaurant you like without being told which one. It finds a dog sitter who fits your schedule. It DJs your playlist based on what you actually listen to, not what you say you listen to.
Key differences from assistant-mode AI:
- Persistent memory across sessions and tasks
- Autonomous execution with its own compute environment
- Proactive task completion, not just response generation
This is the clearest sign yet that Meta sees agents as a business model, not just a product feature. The company has been racing to catch OpenAI and Google in the foundation model wars, but foundation models are expensive to train and hard to monetize directly. Agents that live inside Meta's ecosystem — Instagram, WhatsApp, Messenger, Facebook — and can execute transactions? That's a revenue model. Every restaurant booking, every purchase, every form filled out is a potential commission, a potential ad placement, a potential data point that makes the next recommendation better.
The real competition isn't between Hatch and ChatGPT. It's between Hatch and your browser's autofill. Between Hatch and the delivery app you already use. Between Hatch and the mental overhead of managing your own digital life. If Meta can make Hatch feel like it saves you more time than it costs in setup and oversight, they win a permanent presence in the background of your day.
The Implication
Watch how Meta prices this. If Hatch is free but monetized through transaction fees or "recommended" services, it becomes an agent optimized for Meta's revenue, not your time. If it's a subscription, it competes directly with OpenAI and Google for who gets to be your default digital assistant.
The bigger question: do you want the company that built its empire on algorithmic feeds and attention harvesting to also run the agent that books your dinner, manages your calendar, and "improves your relationships"? The memo's language is telling. Not "helps you book a table" but "improve your life." That's the pitch. Whether it's the reality depends on what Hatch optimizes for when your goals and Meta's incentives don't align.