Meta Prepares Hatch, the AI Agent Inspired by OpenClaw That Shops on Instagram
Meta is building an AI agent aimed at the general public, inspired by OpenClaw and internally dubbed Hatch, along with a tool for automated shopping within Instagram. The goal, according to The Information, is to bring autonomous agents into the hands of the billions of users of the company, also turning them into a commercial lever.
This is not a surprising move: at the quarterly report at the end of April, Mark Zuckerberg already hinted that Meta was working on agents designed for both individuals and businesses. The new report adds names, models, and timelines to that framework.
What is Hatch and What Model Does It Use Today
Hatch is based on the model of OpenClaw, the open-source autonomous agent that went viral at the beginning of the year, capable of executing sequences of actions autonomously by relying on large language models. This is the root of the project: Zuckerberg described OpenClaw as "exciting" but too complicated to set up for most people, and he wants a simpler and more accessible equivalent that can understand the user’s goals and work day and night to achieve them.
The agent is expected to operate both within the group’s apps and with external services. Meta has tested it in closed environments simulating services like DoorDash, Reddit, Etsy, and Outlook, training it under controlled conditions before exposing it to the real world. Among the capabilities the company is reportedly working on is the ability to take initiative autonomously, acting when appropriate instead of merely executing commands.
On the model front, there is a relevant detail: currently, Hatch operates on Anthropic models, specifically Claude Opus 4.6 and Claude Sonnet 4.6. This is a transitional solution because at launch, Meta intends to move the agent to its proprietary model Muse Spark, presented in April. Internal testing is expected by the end of June.
The Challenge to TikTok Shop Within Instagram
The second piece is an integrated shopping agent in Instagram, designed to compete with TikTok Shop. The idea is to allow anyone seeing a product in a Reel to obtain information and complete the purchase without leaving the app, eliminating the steps that currently scatter sales between direct messages to the seller and referrals to external sites. The integration is expected before the fourth quarter.
Meta had already laid the groundwork by allowing creators to tag up to 30 products in a single video. The card the company plays is distribution: rivals offer the agent as a horizontal layer that customers must integrate into their own applications, while Hatch would be born directly within Instagram, which has hundreds of millions of users, without merchants needing to build complex integrations beyond the catalog mapping.
The Race for Personal Agents
Google is also internally testing a twin project called "Remy" and has shut down its previous browser agent, Project Mariner, to concentrate resources on responding to OpenClaw. Further down the line are already Anthropic with Claude Code and Claude Cowork, and OpenAI, which has built on the work of OpenClaw's creator in addition to its own coding agent Codex.
Hatch is partially a fallback: Meta had tried to hire the same creator of OpenClaw, who chose OpenAI instead; and the acquisition of Manus, which we discussed, was later blocked by the Chinese commission NDRC, which ordered its cancellation for national security reasons. In the meantime, Meta's spending on AI infrastructure has risen this year to $145 billion, according to The Decoder.
Meta can count on a significant distribution advantage, which is hard to replicate for those starting from a generalist agent. The key issue in the medium term is another: as long as Muse Spark does not reach production maturity, Instagram's flagship agent will continue to operate on Anthropic models, which are those of a direct competitor.