What is the story about?
Meta is preparing for what could be one of its most significant consumer AI launches yet, with plans to introduce an AI agent internally called Hatch in the coming weeks, according to internal documents reviewed by The Information.
The product is said to be part of CEO Mark Zuckerberg's wider effort to turn Meta's heavy investment in artificial intelligence into new businesses beyond its advertising operation. The company has reportedly considered a tiered subscription structure for the service, with its highest-priced option potentially costing as much as $199.99 a month and offering increased usage limits.
The reported plans come as Meta steps up its efforts in AI agents and advanced models, areas where it is competing with companies including OpenAI and Anthropic.
Early versions of Hatch have been trained to interact with websites and services including DoorDash, Etsy, Reddit, Yelp and Outlook, according to The Information. The idea appears to be to give users a central place from which AI agents can perform different tasks or access specialised tools.
The interface is expected to include a customisable dashboard displaying tools or skills created by agents. Those could range from a fitness tracker to assistance with planning a trip.
Internal documents reportedly place Hatch's launch in late August or early September, although Meta's product timelines could still change. A spokesperson for the company declined to comment on the plans.
Meta is also working on another way to bring AI agents closer to its users. The company is reportedly preparing a platform within WhatsApp that would allow people to connect with and chat to other AI agents through the messaging service.
A limited rollout to a smaller group of users could begin as soon as this week, The Information reported.
Together, the two projects suggest Meta is looking beyond standalone chatbots. The company appears to be building products in which AI systems can interact with external services, perform specific tasks and potentially operate through platforms already used by billions of people.
Alongside the reported Hatch launch, Meta is aiming for an October release of a new AI model known internally as Watermelon, according to the documents. It remains unclear whether the model will join Meta's Muse family or be introduced separately.
Meta has increased the pace of its AI releases in recent months as it attempts to narrow the gap with leading rivals.
In April, the company introduced Muse Spark, its first model developed by Meta Superintelligence Labs. The multimodal reasoning model was designed with tool use and multi-agent systems in mind.
An updated version, Muse Spark 1.1, followed in July, with a greater focus on coding, computer use and agentic capabilities. Meta then released Muse Spark 1.2 in August alongside Muse Code, an AI coding agent aimed at software engineering tasks.
The rapid succession of releases reflects the growing importance of AI agents that can handle more than text generation, including interacting with software and carrying out multi-step tasks.
At the same time, Meta has continued to pursue open models. Earlier this month, it introduced Muse Glimmer, a smaller open model, while Zuckerberg argued separately for wider access to open-source AI. The company has also said it plans to release an open-weight version of Muse Spark 1.2.
That leaves Meta pursuing two tracks at once: developing proprietary products and models capable of competing with closed AI systems, while continuing to make selected models available more openly.
With Hatch reportedly nearing release and Watermelon expected later in the year, Meta's next challenge will be turning that growing pipeline of models and agents into products that can attract users — and eventually generate meaningful revenue.
The product is said to be part of CEO Mark Zuckerberg's wider effort to turn Meta's heavy investment in artificial intelligence into new businesses beyond its advertising operation. The company has reportedly considered a tiered subscription structure for the service, with its highest-priced option potentially costing as much as $199.99 a month and offering increased usage limits.
The reported plans come as Meta steps up its efforts in AI agents and advanced models, areas where it is competing with companies including OpenAI and Anthropic.
What is Meta's new AI agent, Hatch
Early versions of Hatch have been trained to interact with websites and services including DoorDash, Etsy, Reddit, Yelp and Outlook, according to The Information. The idea appears to be to give users a central place from which AI agents can perform different tasks or access specialised tools.
The interface is expected to include a customisable dashboard displaying tools or skills created by agents. Those could range from a fitness tracker to assistance with planning a trip.
Internal documents reportedly place Hatch's launch in late August or early September, although Meta's product timelines could still change. A spokesperson for the company declined to comment on the plans.
Meta is also working on another way to bring AI agents closer to its users. The company is reportedly preparing a platform within WhatsApp that would allow people to connect with and chat to other AI agents through the messaging service.
A limited rollout to a smaller group of users could begin as soon as this week, The Information reported.
Together, the two projects suggest Meta is looking beyond standalone chatbots. The company appears to be building products in which AI systems can interact with external services, perform specific tasks and potentially operate through platforms already used by billions of people.
Watermelon targeted for October as Meta speeds up AI releases
Alongside the reported Hatch launch, Meta is aiming for an October release of a new AI model known internally as Watermelon, according to the documents. It remains unclear whether the model will join Meta's Muse family or be introduced separately.
Meta has increased the pace of its AI releases in recent months as it attempts to narrow the gap with leading rivals.
In April, the company introduced Muse Spark, its first model developed by Meta Superintelligence Labs. The multimodal reasoning model was designed with tool use and multi-agent systems in mind.
An updated version, Muse Spark 1.1, followed in July, with a greater focus on coding, computer use and agentic capabilities. Meta then released Muse Spark 1.2 in August alongside Muse Code, an AI coding agent aimed at software engineering tasks.
The rapid succession of releases reflects the growing importance of AI agents that can handle more than text generation, including interacting with software and carrying out multi-step tasks.
At the same time, Meta has continued to pursue open models. Earlier this month, it introduced Muse Glimmer, a smaller open model, while Zuckerberg argued separately for wider access to open-source AI. The company has also said it plans to release an open-weight version of Muse Spark 1.2.
That leaves Meta pursuing two tracks at once: developing proprietary products and models capable of competing with closed AI systems, while continuing to make selected models available more openly.
With Hatch reportedly nearing release and Watermelon expected later in the year, Meta's next challenge will be turning that growing pipeline of models and agents into products that can attract users — and eventually generate meaningful revenue.
