Meta Platforms Bets on AI Agents: In-House "Hatch" and Instagram Shopping Tool Target TikTok

Meta Platforms is going all-in on AI agent technology, viewing it as the core product form of the next generation, while simultaneously advancing two key fronts internally.

On May 5, tech media outlet The Information reported that sources familiar with the matter revealed that Meta Platforms is developing a consumer-facing AI agent product, 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 stated that these moves are a direct reflection of CEO Mark Zuckerberg's push for an AI agent strategy. During the company's quarterly earnings call last week, he stated that Meta Platforms' goal is to "build agents that understand user goals and help users achieve those goals around the clock."

At the same time, Meta Platforms announced an increase in its AI infrastructure capital expenditure for this year to a maximum of $145 billion. The launch of the Instagram shopping agent is viewed internally as a significant chip in its head-to-head competition with TikTok Shop.

Hatch: From OpenClaw to In-House Agents

The report indicated that the development of Hatch was directly inspired by OpenClaw—a tool that rapidly gained popularity in the tech community for its ability to build autonomous AI agents. Zuckerberg had attempted to acquire OpenClaw, but according to its founder Peter Steinberg in a podcast, the tool was ultimately acquired by OpenAI in February this year, and Meta Platforms' acquisition attempt failed.

In this context, Meta Platforms shifted to independently developing a similar consumer-grade agent product. According to sources, Hatch is currently powered by Anthropic's Claude Opus 4.6 and Claude Sonnet 4.6 models, but will switch to Meta Platforms' latest self-developed AI model, Muse Spark, upon official launch.

To test Hatch's actual capabilities, Meta Platforms specifically built a "sandbox" network environment, simulating real websites such as DoorDash, Etsy, Reddit, Yelp, and Outlook, allowing the agent to train and test under controlled conditions.

Zuckerberg also acknowledged on the earnings call the technical challenges of bringing OpenClaw-like tools to Meta Platforms' billions of users, including heavy reliance on large-scale infrastructure and ensuring the product is simple and easy to use—OpenClaw remains too complex for most non-technical users.

Capability Iteration: Proactivity, Memory, and Tool Invocation

Meta Platforms is currently accelerating the enhancement of Hatch's core capabilities across multiple dimensions.

According to sources, the company is focusing on strengthening Hatch's proactive decision-making abilities, enabling it to take initiative at appropriate times rather than passively waiting for user commands.

At the same time, Meta Platforms is expanding the model's context window and reinforcing its cross-conversation memory capabilities, allowing the agent to continuously remember relevant user information across different sessions. Additionally, the company is optimizing the agent's response methods and the logic for selecting and invoking external tools.

Notably, Meta Platforms already has an AI agent for internal employee use called MyClaw, which is used to access work files, summarize internal company forum posts, and obtain technical advice.

However, as previously reported by The Information, MyClaw recently triggered a major security alert within Meta Platforms—an employee followed incorrect advice given by the agent, resulting in unauthorized access to sensitive company and user data by other employees. This incident highlights the real-world challenges AI agents still face regarding reliability.

Instagram Shopping Agent: Directly Targeting TikTok Shop

On the consumer front, Meta Platforms' other battle line involves deeply embedding AI shopping agents into Instagram.

According to sources, the core functions of this tool include: users can directly click on products in Instagram Reels or feeds to view detailed information, jump to external webpages, and complete purchases within the platform, all without leaving the app.

This feature will be built upon the AI shopping experience upgrades Meta Platforms released in March this year, which included presenting richer product information via AI and a new checkout process allowing users to complete purchases by directly clicking on ads.

Sources also revealed that Meta Platforms hopes to use this tool to increase competitive pressure on TikTok Shop. Leveraging the deep integration of short videos and e-commerce, TikTok Shop has established a significant advantage in the social commerce sector, and Meta Platforms aims to counter this with Instagram's massive user base and AI capabilities.

In the broader competitive landscape, Google launched Gemini Enterprise for Customer Experience in January this year, featuring AI shopping agent functions that support product recommendations, cart building, and automatic checkout after user authorization; Amazon offers the AI shopping assistant Rufus, helping users track prices, research, and purchase goods.

Analysts point out that the advancement of Hatch and the Instagram shopping agent is a microcosm of Meta Platforms seeking commercial returns on its massive capital investments in AI. Zuckerberg positions AI agents as the core of his "personal superintelligence" vision, describing them on the earnings call as round-the-clock tools to help users achieve their goals.

As Meta Platforms raises the cap on its AI infrastructure capital expenditure for this year to $145 billion, market attention continues to intensify regarding when these AI investments will translate into substantial revenue.

Whether Hatch can find a balance between technical complexity and mass usability, and whether the Instagram shopping agent can effectively leverage the social commerce market, will be key indicators testing the success of this strategy.