Meta has opened an internal test of a personal AI agent, code-named "Hatch," to a subset of employees. Unlike a chatbot that answers a question, the agent carries out a chain of concrete actions: filling out a form, booking an appointment, comparing products or completing a purchase on the user's behalf, with every action still subject to approval first.
From an assistant that answers to an agent that acts
To train Hatch ahead of any public rollout, Meta built simulation environments that mirror real sites: DoorDash for food orders, Etsy for handmade goods, Reddit and Yelp for research, Microsoft Outlook for inbox management. Inside these sandboxes, the agent learns to navigate, fill in fields and carry a task through to completion without a human stepping in at every stage.
A shopping push built for Instagram
Beyond administrative tasks, Meta is eyeing a use case tied more directly to payment: a tool that would let a user tap a product spotted in a Reel or a post, with the agent then comparing options and carrying the purchase through without the user ever leaving the app. The stated goal is to cut friction out of mobile shopping and push back against TikTok Shop, which already holds ground there.
A crowded market, still-thin adoption
Hatch isn't alone on this turf. OpenAI has offered an "Instant Checkout" for several months that closes the purchase without leaving ChatGPT, Google is building an "agentic checkout" paired with merchant inventory checks, and Perplexity relaunched its own shopping tool with sharper personalization. Meta is running a second, already-public initiative, separate from Hatch: a product-recommendation carousel built into the Meta AI chatbot, tested with select US users, showing image, brand, price and a link to the merchant's site, with no built-in checkout.
| Player | Feature | Built-in checkout |
|---|---|---|
| OpenAI | Instant Checkout inside ChatGPT | Yes |
| Agentic checkout with inventory verification | Yes | |
| Meta | Meta AI chatbot carousel (in test) | No, links out to the merchant |
| Meta | Hatch (internal employee test) | Planned, not yet public |
The gap between the ambition the major platforms are advertising and the actual usage EMARKETER measured, barely 2% of US digital buyers, puts Hatch in a betting phase rather than a catching-up one. The contest isn't playing out on volume yet, but on whether the experience, recommendation accuracy, product availability, and a reliably closed transaction, gets good enough that users actually hand the purchase over instead of going back to a manual click.