WhatsApp adds up to five third‑party AI agents

WhatsApp opens its chat to external AI agents

WhatsApp announced a feature that lets users attach up to five third‑party AI agents to a conversation. The agents appear as separate contacts and can be instructed to perform tasks, from drafting messages to booking appointments.

The rollout is still in testing, but the plan is to expose a new API that developers can hook into. WhatsApp will surface the agents inside the existing chat UI, allowing users to switch between them without leaving the app. The move puts WhatsApp in direct competition with other messengers that already host bots, and it raises questions about how open‑ended AI will be governed inside a platform that processes billions of daily messages.

Autonomous agents in practice

The promise of “agents that act for you” is no longer theoretical. A recent benchmark from Bottleneck Labs ran seven autonomous AI businesses for a month. Those agents generated $12,431 in fake invoices and lost $3,200 in real‑world costs. The experiment showed that language models can execute end‑to‑end workflows, but it also exposed financial leakage and the need for tight supervision.

WhatsApp’s integration could enable similar use cases at consumer scale. A user might ask an AI to reorder groceries, and the agent could place the order through a partner service without the user typing anything. The same capability could be abused to send spam or phishing messages that appear to come from a trusted contact. The platform will have to decide how much autonomy to grant each agent and what safeguards to enforce.

Patent friction and the risk of over‑protection

The Electronic Frontier Foundation recently highlighted a patent owned by food‑tech company Hampton Creek that claims a “machine‑learning enabled discovery platform.” The patent lists every standard technique taught in an introductory AI class—linear regression, k‑nearest neighbors, decision trees, and more—applied to protein data. EFF argues that such breadth stifles innovation because any future product that uses those methods on a new dataset could be accused of infringement.

If WhatsApp’s AI API becomes a standard way to embed third‑party agents, developers may run into similar patent thickets. A developer could be forced to license a broad claim that merely covers “using machine learning to process user messages.” The EFF analysis suggests that the patent office is increasingly willing to grant patents on obvious applications of known algorithms, a trend that could slow the rollout of useful agents or inflate licensing costs.

Industry context and competitive pressure

Messaging platforms have long experimented with bots. Slack opened its bot ecosystem in 2015, and Telegram introduced a bot API that now powers thousands of services. Those ecosystems grew because the platforms offered clear guidelines and revenue‑sharing models. WhatsApp, by contrast, has kept its API tightly controlled for business accounts, primarily to prevent spam.

Opening the consumer side to AI agents signals a shift in strategy. The move may be driven by user demand for in‑app assistance and by competitors that already embed AI assistants in their native keyboards or voice layers. It also aligns with a broader trend of “AI‑as‑a‑service” where developers package models as plug‑ins rather than building monolithic systems. The key difference is that WhatsApp’s user base is massive and highly regulated in terms of privacy, which could make the integration riskier.

What to watch

WhatsApp will start a limited rollout of the agent API in the next few weeks. Watch for the first set of approved third‑party agents and the policy documents that accompany them. The next data point will be whether the platform publishes usage metrics or reports of abuse. Parallel to that, monitor any patent litigation that cites the Hampton Creek filing or similar broad AI claims, as those cases could set precedents for how much freedom developers have when building agents for WhatsApp.