Meta tests agentic AI features in WhatsApp as its Hatch platform and “Watermelon” model move toward consumer monetization

Meta is running limited trials that extend its consumer AI strategy beyond chatbot-style conversations, with tests aimed at integrating more agentic, third-party AI experiences directly into WhatsApp. The approach, described in a report by Yahoo Finance, centers on Meta’s Hatch Agent Platform and a model referred to as “Watermelon.”

The reported direction matters because it changes how users would interact with AI inside one of Meta’s most used apps. Instead of prompting a system to generate text, “agent” style software is designed to carry out multi-step tasks, using tools and external services as needed. Embedding those capabilities into WhatsApp would put AI assistance closer to the messaging flow, where users already request help, share context, and coordinate actions.

Yahoo Finance also frames the trials as part of a broader monetization effort. In that view, Meta’s goal is not only to improve engagement, but to make AI-generated services easier to trigger inside WhatsApp, including third-party experiences, which could become targets for paid distribution, subscriptions, or commerce-related payments.

While the report emphasizes experimentation, it does not announcement a broad rollout. The trial framing suggests Meta is still assessing product fit, reliability, and user demand before scaling any consumer-facing agent experiences. That kind of phased release is common in AI products where safety controls, content moderation, and tool access have to work reliably across many real-world messaging scenarios.

Meta’s platform approach through Hatch is positioned as an enabler, not just a feature. A platform strategy typically means developers can connect services or workflows to an AI layer, rather than Meta building every possible agent capability itself. In practical terms, if WhatsApp can host agent workflows sourced from partners, Meta can expand the range of tasks available to users without needing to train and maintain a single monolithic assistant for everything.

The “Watermelon” model reference, as described in the report, points to an internal or curated model capability that Meta is using for these consumer experiments. Model naming aside, the central question for users and partners is performance under constraints, such as responding appropriately in short-form messaging, handling ambiguous requests, and managing multi-step task execution without producing misleading or ungrounded outcomes.

Meta has not disclosed, in the limited information available from the reported account, key rollout details such as where the trials are running, what specific third-party agent experiences are included, what pricing or commercialization mechanics are being tested, or what safeguards are in place for tool-using agents inside WhatsApp.

For investors and product watchers, the next indicates to watch are whether Meta broadens the trial footprint, adds more partner-powered agent use cases, and clarifies how monetization would work in practice, for example through commerce flows, premium tiers, or developer/partner arrangements. Equally important will be any public description of safety and policy controls for agent actions that occur inside everyday messaging, where errors can propagate quickly.

Meta said in its official newsroom that it continues to advance AI across products and infrastructure, but the specific Hatch and “Watermelon” trial details referenced by the Yahoo Finance report are not confirmed in the newsroom page itself. As with many early AI product tests, the current public picture is directional and incomplete, and more detail would likely come through additional product announcements or formal documentation as trials mature.