Meta (NASDAQ:META) is said to be internally testing a consumer AI agent designed to complete everyday work on the open web, according to an employee memo described by BI. Internally labeled Hatch — a working name, not the expected public brand — the system is being positioned as more than a chatbot.
Staff documents say it has access to its own computing environment so it can fill out forms, make purchases, run in-depth research, and otherwise handle activities a person could perform in a browser.
The trial has already moved beyond Meta’s core AI laboratory. Reporting says the tool was opened last month to employees outside that lab, giving a wider group of staff a chance to use it on ordinary personal errands.
One product manager described an agent named Veda that helped locate a dog sitter, buy a Father’s Day present, and adjust sleep routines.
Those examples illustrate the intended use case: a persistent helper that can follow multi-step requests rather than only answering questions.
Company notes distinguish Hatch from a conventional assistant by emphasizing memory, customization, and action.
The memo claims the agent retains more context than typical models and can connect to services such as email, calendars, Spotify, Instagram, and OpenTable.
Users would name the agent, shape its tone, and decide what it should watch. If it errs, they can correct it and expect it to adapt.
For higher-risk steps — payments, bookings, or other sensitive operations — the design calls for explicit approval before the agent proceeds.
Earlier coverage of the same program adds that prototypes have been trained to work with sites including DoorDash, Etsy, Reddit, Yelp, and Microsoft Outlook.
Early interfaces reportedly include a dashboard of reusable “skills,” such as a fitness tracker or a travel itinerary builder.
That architecture would let people treat the agent as a personal workspace rather than a one-off chat. Meta has also been reported to be weighing paid tiers, including a high-end option that could reach about $200 a month with higher usage limits.
Launch timing and final pricing remain unconfirmed, and a company spokesperson declined to comment.
The experiment fits a broader industry shift from conversational AI to agentic systems that can click, type, shop, and schedule.
Meta already offers planning features in Meta AI and a Business Agent for WhatsApp and Instagram customer workflows.
Hatch appears aimed at individual consumers and at converting the company’s large AI spend into a product people might subscribe to, alongside ads, models, and developer tools.
Mark Zuckerberg has publicly argued that polished personal agents — easier to use than raw open-source agent stacks — are a major opportunity.
Reliability and permissioning will decide whether that opportunity is real.
Giving software a computer and credentials raises familiar questions about mistaken purchases, privacy, and third-party site policies.
Meta’s reported approval gates are an attempt to keep a person in the loop. Until a public release, Hatch remains an internal trial: a glimpse of how Meta wants assistants to stop describing the web and start working on it.