AI news ·
Meta develops computer-operating AI agent Hatch using Anthropic models
Meta is building an AI agent called Hatch that can control computers and handle tasks like shopping and payments. Internal testing wraps up next month, with a planned switch to Meta's own Muse Spark model before release.
Meta Develops Computer-Operating AI Agent Called Hatch
Meta is building an AI agent designed to directly control users' computers and complete tasks like shopping, payments, and document work. The company, which calls the agent Hatch, plans to finish internal testing next month, according to The Information.
Hatch draws inspiration from OpenClaw, a similar system that gained attention after its November launch. Meta is currently training Hatch in simulated environments modeled on DoorDash and Yelp to teach it how to navigate real-world applications.
Current and Future Models
Meta is building Hatch using Anthropic's Opus 4.6 and Sonnet 4.6 models. By the time of release, the company plans to replace these with its own model, Muse Spark.
Strategic Positioning
CEO Mark Zuckerberg has positioned AI Agents & Automation as central to Meta's vision of "personalized superintelligence." On the company's April earnings call, he described agents that understand user goals and work continuously to achieve them.
Meta attempted to recruit Peter Steinberg, an OpenClaw developer, earlier this year but was unsuccessful. In March, the company acquired Moltbook, a social media platform used by OpenClaw developers, as an alternative approach to building expertise in this space.
For IT teams and developers, understanding how AI for IT & Development operates at the system level will likely become relevant as these agents move into production environments.