What was announced
Sierra, the customer service agent company founded by Bret Taylor and Clay Bavor, announced on October 6 that it is developing, together with Meta and several partners, the Personal Agent Protocol, an open standard that anyone can implement. The initial partners include Genesys, Instinct, Rocket, Shopify, Stripe and Walmart. According to the summary of Meta's announcement published by Relevant Audience, Meta is developing the protocol based on its personal agent Muse and on Meta Business Agent, and also mentions NiCE and Decagon.
On October 9 Sierra added 35 design partners, including Visa, Mastercard, PayPal, OpenAI, Target and United Airlines, and published a draft of the specification at personalagentprotocol.org. The plan includes design workshops and a reference implementation for developers.
How the protocol would work
Today, when an AI agent wants to carry out a task for a person, it browses web pages as a human would and often ends up on the customer service line or in the web chat. The idea of the protocol is to make the connection direct and secure. According to Sierra:
- Discovery: the agent starts on the company's website, where it finds what the company offers and how to connect.
- Authentication: it can come in as a guest (for example, to check availability) or log in on the customer's behalf with OAuth, a well-known sign-in standard.
- Permissions: the customer decides whether their agent can only look things up or also make changes, and the company decides what it makes available.
- Channels: the company chooses whether the agent uses its website, its APIs (with standards such as MCP or OpenAPI) or the company's own AI agent, for tasks that require a conversation, such as a warranty claim.
Sierra mentioned notifications (for example, flight or shipping delays) and payments without sharing card details as future extensions.
The context: Meta Business Agent and Muse
Meta introduced Meta Business Agent on June 3, 2026 at its Conversations event in London, and extended it to businesses of all sizes worldwide. According to the official WhatsApp Business blog, the agent answers questions about the business, recommends catalog products, books appointments, qualifies leads and closes sales, in the customer's language and with the brand's tone; the company decides when a person steps in. Meta says more than one million businesses were already using it on WhatsApp and Messenger. For large enterprises there is Meta Business Agent Platform, which connects to systems such as Shopify and Zendesk.
Getting started is free in the app, with paid subscription plans. On the API, since August 1, 2026 Meta has charged for agent messages at 2 US dollars per million tokens, which according to its documentation typically works out to about 4 to 5 US cents per message. On the other side is Muse, the personal agent Meta launched in September in the United States so consumers can book, buy or handle requests. The new protocol is the bridge between both worlds: the customer's agent and the company's systems or agent.
Meta's agent or agents connected to your phone system and CRM
For a company, the practical question is where its AI agent lives. Based on what has been published so far, these are the main differences:
- Channels. Meta Business Agent covers WhatsApp, Messenger and Instagram. Meta has not announced that it handles phone calls, which remain a key channel in Latin America.
- Integration. Meta's agent connects to the systems Meta supports. Your own agent can directly query and update your CRM, inventory or scheduling system.
- Cost. With Meta, AI and message delivery are charged together by tokens. With a third-party agent, the company pays the AI provider and, separately, pays Meta for the messages.
- Control. With your own agent, you define how conversations are stored, how they are transferred to a human agent and how you comply with Ley 1581 de 2012, Colombia's data protection law.
What it means for companies in Colombia and Latin America
The protocol is still a preview and Muse is not available in the region, so there is nothing to change tomorrow. But the direction is clear: customers will not only write or call; in some cases their AI agent will. Companies that have their information in order (prices, opening hours, order status) and processes accessible via API will be better prepared.
The advisable step today is to define which tasks an AI agent can handle, when it should hand over to a person and which channels it should cover: chat, WhatsApp and phone. At 19 80 TIC Cloud we deploy voice and chat AI agents connected to your phone system, to WhatsApp and to the QTbrain CRM, with transfer to human agents when the case calls for it.
