Meta presents a personal agent capable, according to the company, of taking steps and continuing work after the application is closed. Its adoption will depend as much on trust as on its capabilities.
Preparing a reservation, filling out a form or organizing a project: with Muse, announced on September 8, Meta wants to develop the delegation of tasks to artificial intelligence.
The agent is being deployed first in the United States, on mobile and web. Users can also interact with it in WhatsApp. Meta indicates that it can continue working in the background and return to the user when validation is required. Source: Meta announcement, September 8, 2026.
Assigning a task implies assigning access
To manage these actions, Meta describes a dedicated virtual computer for each user and a separate system, Sentinel, which controls internet communications. The company announces validation before sensitive operations, such as purchases or sending emails, as well as an activity log.
It also states that conversations and data from this environment are not shared with its advertising systems. These are the guarantees announced by Meta; this article does not constitute an audit of their effectiveness. Source: Muse official presentation.
The real test: how much time is saved?
Our analysis: a personal assistant becomes useful when it reduces the amount of supervision required. If every step needs to be checked, choices corrected, and procedures repeated, the benefit quickly diminishes.
The trials will therefore need to measure simple results: task completed, information accurate, cost respected, and human intervention limited to the right decisions.
Muse also raises a question for brands: how to present an offer when an agent is involved in the purchasing process? Clarity of prices, conditions and availability could become as crucial as the power of advertising.
AI-generated illustration — Wazup-InTech.
