Via pixabay.com
The social media giant is building a web-navigating AI assistant trained on simulated versions of DoorDash, Etsy, and other services, marking its first serious push into paid AI products
Meta is preparing to roll out an AI agent called Hatch that can browse the web and complete tasks on your behalf, with a premium subscription tier priced at $199.99 per month. The product represents Meta’s first major foray into paid AI services, a significant departure for a company that has historically made its money through advertising.
The agent is currently being trained inside simulated environments that mimic popular services including DoorDash, Etsy, Reddit, Yelp, and Outlook. Internal testing is targeted for June 2026, with a potential public rollout as early as July.
What Hatch actually does #
The premium tier, reportedly called “Hatch Plus,” would offer 5-10x more daily task capacity than a potential free version. That pricing puts it in the same neighborhood as high-end AI subscriptions from OpenAI and Anthropic.
Meta is currently training Hatch using Anthropic’s Claude models rather than its own. The plan is to eventually transition to Meta’s proprietary models.
By building controlled mock environments, Meta can train the agent in conditions it can actually measure and iterate on before unleashing it on the messy reality of the actual internet.
Why Meta wants your subscription money #
Mark Zuckerberg has been vocal about the need to diversify Meta’s revenue model beyond advertising. The company has poured billions into AI infrastructure, and while that investment powers the recommendation algorithms that keep users glued to Instagram and Facebook, it hasn’t yet produced a standalone product that generates direct consumer revenue at scale.
The project is also connected to agentic shopping features planned for Instagram, which could enhance the platform’s e-commerce capabilities.
Internally, the initiative is being compared to Meta’s existing Llama model family, which the company has positioned as an open-source alternative to proprietary models from competitors.
The competitive landscape #
Meta is entering an increasingly crowded field. OpenAI has been building agentic capabilities into ChatGPT, and Anthropic has been developing similar features for Claude. Google has its own agent-oriented projects in the works.
What Meta brings to this fight is distribution. If Hatch gets integrated into Instagram, Facebook, WhatsApp, and Messenger, the adoption curve could look very different from a standalone app that needs to acquire users from scratch.
There are real risks. An AI agent that buys the wrong thing on DoorDash is annoying. One that sends the wrong email through Outlook is potentially career-ending. The reliability bar for agentic AI is significantly higher than for chatbots, because the consequences of errors are real and immediate. Meta’s decision to train in simulated environments before going live suggests the company is aware of these stakes.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our