Hatch: Meta Prepares to Launch Its A.I. Agent for the Masses as Early as September

Meta is reportedly getting ready to ship a consumer A.I. agent called Hatch, with a launch window of late August or early September. The account comes from The Information, which cites internal documents, and other outlets have followed it. Meta has not confirmed the name, the feature set, the timing or the price.

That makes Hatch a much bigger bet than the chat assistant Meta AI that people already see. Meta AI mostly responds inside a conversation. Hatch is said to be built to take a goal, figure out the steps, work across connected services and keep going until the task is done. Think less chatbot, more digital gofer with a memory.

The reported tests tell you where Meta thinks this thing should live. According to the reports, the agent has been trained inside simulated versions of DoorDash, Etsy, Reddit, Yelp and Microsoft Outlook. Early prototypes are said to include a customizable dashboard and small tools the agent can assemble itself, like a fitness tracker or a travel itinerary. The list of possible abilities is broad: product research, email drafting, summaries, recommendations, multi-step work across apps, long-running goals and confirmation before purchases, bookings, messages or account changes.

None of that is confirmed, and plenty still hinges on connectors. An agent only becomes useful when it can move between services, and that is also where the permission mess starts. The picture is still murky on which interfaces Hatch would use and which partners would be ready at launch. It also appears to be separate from Meta’s shopping agent for Instagram, which is a commerce feature rather than a broad consumer helper.

The price is the other headline-grabber. Internal documents reportedly show Meta considering several tiers, with the top plan reaching $199.99 a month for higher usage limits. That would sit well above the Meta One subscriptions the company has tested in Singapore and Guatemala, and it would look steep next to the top plans from OpenAI, Anthropic and Google. Meta already charges by usage for its WhatsApp Business Agent, so the expensive part is not the concept. It’s the compute bill behind agents that can spend minutes moving around the web.

What happens under the hood may matter as much as the product itself. Muse Spark looks like the likeliest launch base, with Muse Code, Muse Glimmer and Watermelon also in the frame depending on task complexity and pricing tier. The shape that emerges sounds less like one giant model and more like routing: cheap models for simple work, bigger ones for the hard parts. That is the part most companies eventually discover after the hype smoke clears.