Meta’s Project Hatch AI Agent: What It Could Do, When It May Launch and What It Could Cost

Meta’s Project Hatch is reportedly being developed as a consumer personal AI agent—a system that can carry out online tasks instead of only responding to prompts. Reports describe an employee-tested product that could research options, navigate websites, fill out forms, make purchases and continue working on a user’s behalf in the background. 1718

The important caveat is that Hatch is not a publicly launched product with a confirmed feature list. The available information comes from reports citing internal documents and employee communications, so its name, integrations, launch timing, permissions and price may change.

What Project Hatch is expected to do

Hatch is described as a more action-oriented alternative to a conventional chatbot. A user could state a goal in everyday language, and the agent would reportedly plan the steps, use connected services and ask for approval when an action is sensitive. 1819

Reported examples include:

• Booking a restaurant table
• Ordering food
• Shopping and buying gifts
• Finding a dog sitter
• Filling out online forms
• Making purchases
• Conducting detailed research
• Managing information related to finances and relationships
• Helping users work on habits such as sleep
• Controlling music or acting as a DJ through Spotify

These examples describe the product’s reported ambitions, not guaranteed capabilities in a final consumer release. The strongest evidence so far concerns online errands, research, shopping and connections to common consumer services. 171820

Reported integrations: email, calendars and commerce services

The reported integration list spans personal information, social platforms, entertainment and online commerce. Examples include:

• Email and calendars: to use schedule and message context when planning tasks
• Instagram: to work with information inside Meta’s social ecosystem
• Spotify: to control music and create a more personalized listening experience
• OpenTable: to research or book restaurant reservations
• DoorDash: to order food
• Etsy: to browse or purchase gifts and other products
• Reddit and Yelp: to research recommendations and reviews
• Microsoft Outlook: to work with email and calendar information

Some reports list only the initial examples, while others describe a broader set of planned connections. That means the complete integration roster—and whether each service will support reading, research, booking or purchasing—has not been established. 41518

The proposed architecture is significant because it would combine context access with computer control. Instead of giving a user a restaurant recommendation and stopping there, an agent could potentially check the calendar, compare options, make a reservation and return with the result. That is the promise of agentic software; it is also where mistakes become more consequential.

How users may customize Hatch

Reports portray Hatch as a highly personalized agent rather than a fixed assistant. Users may reportedly be able to:

• Name the agent
• Choose its conversational personality
• Decide what it pays attention to
• Set how proactive or attentive it should be
• Define which accounts and services it can access
• Choose which actions require approval

A reported memo says users would be able to approve sensitive actions before they happen. However, the available reporting does not establish the final permission screen, account-isolation model, spending limits, audit history or security safeguards. 20

That distinction matters. Reading a calendar is different from sending an email; researching a product is different from buying it; and suggesting a restaurant is different from making a reservation. A usable consumer agent will need clear boundaries between those levels of authority.

Hatch launch timing and possible price

As of August 29, 2026, Hatch was reportedly being tested with Meta employees. Reports have variously pointed to a consumer release in the coming weeks, potentially in late August or early September, but Meta has not publicly committed to a final launch date. 691217

Meta has also reportedly considered a tiered subscription model. The highest tier could cost as much as $199.99 per month, according to reports. That figure is a contemplated price, not an announced subscription or confirmed feature bundle. 45811

The possible price suggests Meta may position the most capable version as a premium service for users who want heavier usage or more demanding autonomous tasks. It does not establish whether a free tier will launch, what lower tiers would include or how transactions would be billed.

What Watermelon could have to do with Hatch

Reports say Meta is targeting October for a new AI model codenamed Watermelon, which could improve the reasoning and agentic capabilities behind products such as Hatch. The relationship between the model and Hatch has not been officially specified, so it is not yet clear whether Watermelon would power Hatch at launch or arrive as a separate release. 469

For an agent that works across websites and apps, model quality is only one part of the system. Reliable performance would also depend on the browser or computer-control layer, integrations, identity and authorization systems, transaction handling and the ability to recover when a website changes or an action fails.

How Hatch fits Meta’s broader AI strategy

Meta has publicly described a goal of distributing “personal superintelligence” broadly rather than concentrating it in a small number of institutions. Its vision includes personal agents that understand users’ goals and help across areas of daily life.

Hatch appears to be a practical consumer expression of that strategy: a persistent assistant that could operate across communication, social discovery, entertainment and commerce. Meta’s potential distribution advantage is its existing network of Instagram, Facebook, WhatsApp and Messenger, although a full Hatch rollout across those products has not been confirmed. 101419

Meta’s infrastructure spending reinforces the scale of the bet. The company’s reported 2026 capital-expenditure outlook is $130 billion to $145 billion, with much of the investment directed toward AI infrastructure. That spending could provide capacity for training and running large, frequently available AI systems, but it does not by itself prove that Hatch will launch successfully or achieve the capabilities described in reports. 1

How Hatch compares with other computer-controlling AI agents

Hatch would enter an increasingly competitive category. Google, OpenAI and Anthropic have all been associated with systems that can observe a browser or computer interface and perform actions such as clicking, typing, navigating and completing forms.

The reported distinction for Hatch is less about inventing computer use and more about combining it with Meta’s consumer distribution and personal context:

• Google: Its agent efforts are closely tied to Gemini and Google’s browser and service ecosystem.
• OpenAI: Operator and ChatGPT’s agent-style capabilities focus on carrying out web tasks for consumers.
• Anthropic: Claude’s computer-use approach is strongly associated with developer workflows and model-controlled desktop interaction.
• Meta: Hatch is reportedly aimed at lifestyle, commerce, relationships and personal routines, with possible access to Meta’s messaging and social surfaces.

These comparisons should be treated as high-level positioning, not a definitive performance ranking. Hatch has not been publicly released in a form that allows independent testing against competing agents.

The central question is whether users will trust it

An agent that can access email and calendars, interpret social context, research products and make purchases has far more responsibility than a chatbot that generates text. The most important launch questions will be practical:

• Can users grant narrowly defined permissions?
• Are purchases and messages paused for approval?
• Can users set spending limits?
• Is there a complete activity log?
• Can access be revoked instantly?
• How does the system handle incorrect memories or conflicting instructions?
• What happens when a website, payment flow or connected account changes?

The reports establish Meta’s ambition, but not yet a demonstrated, independently tested implementation of those protections. Until Meta confirms the product and publishes its controls, Hatch is best understood as a promising but unverified plan for a consumer agent that could move from answering questions to handling everyday digital errands.