Meta is fully betting on AI agent technology, viewing it as the next-generation core product form, and is simultaneously advancing two key initiatives internally.
On May 5, according to tech media The Information, sources revealed that Meta is developing a consumer-facing AI agent product, internally codenamed "Hatch," with the goal of completing internal testing by the end of June. The company also plans to integrate a standalone AI shopping agent tool into Instagram and aims to launch it before the fourth quarter of this year.
The report states that the above strategy is a direct reflection of CEO Mark Zuckerberg's strong push for AI agent initiatives. Last week, during the company’s quarterly earnings call, he said Meta’s goal is to "build agents that understand users’ goals and help them achieve those goals around the clock."
Meanwhile, Meta announced an increase in its 2024 AI infrastructure capital expenditure to as much as $145 billion. The launch of Instagram Shopping Agents is viewed internally as a key competitive move against TikTok Shop.
Hatch: From OpenClaw to Self-Developed Proxy
Reports indicate that Hatch's development was directly inspired by OpenClaw, a tool that gained rapid popularity in tech circles for its ability to build autonomous AI agents. Zuckerberg attempted to acquire OpenClaw, but according to its founder, Peter Steinberg, on a podcast, the tool was ultimately acquired by OpenAI in February of this year, and Meta’s acquisition attempt was unsuccessful.
In this context, Meta has shifted to developing its own consumer-grade proxy product with similar functionality. According to insiders, Hatch is currently powered by Anthropic’s Claude Opus 4.6 and Claude Sonnet 4.6 models, but will switch to Meta’s newly developed AI model, Muse Spark, upon its official launch.
To test Hatch's real-world capabilities, Meta built a dedicated "sandbox" network environment that simulates real websites such as DoorDash, Etsy, Reddit, Yelp, and Outlook, allowing agents to be trained and tested under controlled conditions.
Zuckerberg also acknowledged on the earnings call the technical challenges of bringing OpenClaw-like tools to Meta’s billions of users, including heavy reliance on large-scale infrastructure and ensuring the product is simple and easy to use—OpenClaw is still too complex for most non-technical users.
Capability iteration: Proactivity, memory, and tool invocation
Meta is currently accelerating the enhancement of Hatch's core capabilities across multiple dimensions.
According to reports, insiders said the company is working to enhance Hatch's ability to make proactive decisions, enabling it to take action autonomously when appropriate, rather than passively waiting for user instructions.
Meanwhile, Meta is expanding the model’s context window and enhancing its cross-conversation memory capabilities, enabling agents to retain relevant user information across different sessions. Additionally, the company is optimizing how agents generate responses and their logic for selecting and invoking external tools.
Notably, Meta has an internal AI agent called MyClaw, used by employees to access work files, aggregate posts from internal forums, and obtain technical advice.
However, according to prior reporting by The Information, MyClaw recently triggered a major security alert within Meta—an employee followed incorrect advice from the agent, resulting in sensitive company and user data being accessed by unauthorized staff. This incident highlights the real-world challenges AI agents still face in terms of reliability.
Instagram shopping agent: Direct access to TikTok Shop
On the consumer side, Meta’s other front is deeply integrating AI shopping assistants into Instagram.
According to reports, insiders revealed that the tool’s core features include: users can directly click on products in Instagram Reels or the feed to view details, navigate to external webpages, and complete purchases within the platform without leaving the app.
This feature will build upon the AI-powered shopping experience upgrades Meta released in March, which include richer product information presented via AI and a new checkout flow that allows users to make purchases directly by clicking on ads.
Insiders also revealed that Meta hopes to use this tool to increase competitive pressure on TikTok Shop. TikTok Shop has established a significant advantage in social commerce through the deep integration of short videos and e-commerce, and Meta aims to counter this with Instagram’s large user base and AI capabilities.
In the broader competitive landscape, Google launched Gemini Enterprise for Customer Experience in January this year, featuring AI shopping agents that support product recommendations, cart building, and automated checkout after user authorization; Amazon offers the AI shopping assistant Rufus to help users track prices, research, and purchase products.
Analysis indicates that the advancement of Hatch and Instagram shopping agents reflects Meta’s large-scale capital investment in AI seeking commercial returns. Zuckerberg has positioned AI agents as central to his vision of "personal superintelligence," describing them during the earnings call as round-the-clock tools to help users achieve their goals.
As Meta raises its 2024 AI infrastructure capital expenditure cap to $145 billion, market interest in when its AI investments will translate into tangible revenue continues to grow.
Whether Hatch can strike a balance between technical complexity and大众可用性, and whether Instagram shopping agents can effectively unlock the social commerce market, will be key indicators of the success of this strategy.