Technology companies have released a new class of "agentic" artificial intelligence designed to perform real-world tasks such as managing email, filling out forms, and monitoring websites for product promotions. Unlike standard chatbots that primarily answer questions, these AI agents are intended to act as autonomous assistants. Notable entries in the market include OpenAI’s Dots, xAI’s Grok Bot, and Meta’s Muse, which data analytics firm Sensor Tower identified as one of the fastest-growing applications.
The shift toward autonomous agents follows the development of large language models that power chatbots. While chatbots act as a guide for users who remain in control, agents are compared to self-driving cars where the software executes the tasks under human supervision. Users utilize these tools to handle time-consuming tasks, such as comparing car insurance rates or identifying forgotten recurring charges. For instance, Muse user Scott Burka reported saving approximately $100 per month by using the agent to identify and cancel subscriptions his children had started on their mobile devices.
The cost of these services varies by provider. OpenAI charges approximately $100 per month for a Dots subscription, while xAI’s Grok Bot starts at $30 per month. Meta offers both a free and a paid version of Muse, with some users reporting that the free tier meets their needs. However, the functionality of these agents depends on the amount of personal data a user provides, which can include access to email, calendars, contacts, and Google Drive accounts.
Security and reliability remain central points of discussion. Meta stated that Muse is designed with agents that do not link to one another and limited internet access. The company noted that users maintain control over account permissions and can require the AI to seek approval before performing sensitive actions like buying products. Conversely, Jiaxin Pei, an assistant professor at the University of Texas, Austin, stated that these agents are not yet as reliable as humans and can make mistakes when interacting with third parties.
The concrete change for a person is a shift from manual searching to automated monitoring of accounts and retailers. A user might notice budgeting tips but must also contend with the risk of "dumb shopping mistakes," such as the AI recommending incorrectly sized hardware. While the software can view bank accounts to provide advice if given permission, it currently cannot move funds. The knock-on effect involves a lack of legal clarity regarding liability; if an AI-driven transaction leads to a loss, banks warn it is currently unclear whether the financial institution, the AI developer, or the consumer is responsible for the cost.
What happens next: As more banks and technology firms integrate these agents, the industry is navigating the risks of fraud and scams associated with autonomous software. Financial institutions are calling for the establishment of proper guidelines to determine liability for AI-related errors. There are no reported dates for when such regulations might be finalized or when additional features, such as the ability for agents to move money directly, might be enabled. For now, experts recommend that users double-check the work of these agents before finalizing purchases.