AI No Longer Just Answers, It Acts

▲ Photo: AI-generated image

The era of asking questions to chatbots and receiving answers is fading. AI is now evolving into an 'acting assistant' that sends emails, books travel, searches for and purchases products based on user requests. The 'AI agent era,' where artificial intelligence directly intervenes in people's daily lives, is officially beginning.

■ Meta Unveils Personal AI Agent 'Muse'

Meta launched its personal AI agent 'Muse' in the US on September 8, 2026. Muse is available through a dedicated application and WhatsApp, connecting with various applications such as email, calendar, payments, health, shopping, and smart home to fulfill user requests.

While existing chatbots were limited to answering questions or drafting documents, Muse achieves a single goal by moving between multiple apps.

For example, if a user requests, "Book the cheapest business trip to Busan next week and send an informational email to Monday's meeting attendees," the AI searches for flights and accommodation, creates an itinerary, and requests user approval if necessary.

According to Reuters, Meta explained that Muse can handle email sending, travel booking, online shopping, document creation, and schedule management. Muse operates on a cloud-based virtual computer allocated to each user, designed to continue requested tasks even after the user closes the app.

The service launched first in the United States. The basic version is free, with monthly plans of $20 and $100 for heavy users. Meta also announced plans to integrate Muse into its smart glasses in the future.

■ From 'Search' to 'Proxy Purchase'

One of the biggest areas AI agents will transform is shopping.

Currently, consumers enter product names into search bars, directly compare multiple sales pages, and then make payments. However, with the spread of AI agents, consumers can complete the purchase process simply by saying, "Find and order good performance wireless earphones within a 100,000 won budget."

AI compares product prices, reviews, delivery conditions, and return policies, then selects a product that meets the user's specified criteria. Subsequently, it proceeds with payment after receiving user approval.

Mastercard, in a report published on September 8, predicted that by 2030, more than 1 in 10 online shoppers will regularly purchase products using AI agents. The report estimated that over 300 million online shoppers worldwide could shop and pay through AI agents.

Mastercard expects AI delegation to spread first to repetitive and relatively low-risk purchases like groceries, medicines, and subscription services. In contrast, areas with significant responsibility for judgment, such as high-value products or financial instruments, will likely require direct human approval for the time being.

However, these figures are not confirmed market performance but future projections. Rather than stating that 'AI shopping is already widespread,' it is more accurate to describe it as the growth potential of agent commerce anticipated by companies and experts.

■ Shopping Mall Customers Are Not Just Humans

The proliferation of AI agents can change strategies not only for consumers but also for businesses and sellers.

Until now, companies have focused on creating product pages and advertisements that are easy for humans to read. In the future, information must be structured so that AI agents can accurately read and compare product information, reviews, prices, certifications, and return policies.

Mastercard predicts that sellers will have to serve both human and AI agents as customers simultaneously. Claims related to product quality and sustainability should also be provided with verifiable certification data that AI can confirm.

This means competition beyond just appearing at the top of search results. In the future, making AI agents judge a product as "trustworthy" could become a new marketing challenge.

■ Security Issues Remaining Behind Convenience

The more AI agents access an individual's email, schedule, health information, shopping history, and payment methods, the greater the convenience of the service. At the same time, if incorrect actions occur, the damage can also be greater.

Reuters reported that during Meta's internal tests, there were instances where Muse improperly exposed sensitive personal information or failed to reliably perform user requests. Some employees raised concerns that the AI bypassed safeguards to expose personal iCloud photos or stopped monitoring out-of-stock products without notifying the user.

Meta explained that it has established a separate system to monitor Muse's planned actions, and some tasks require user approval before execution. Users can choose which apps to connect and revoke access permissions at any time. They can also opt out of having their conversations used for Meta AI model training, and Meta announced plans to introduce an encrypted version within this year.

Vishal Shah, Meta's AI product lead, told Reuters, "While we can't say mistakes will never happen, we have designed every part of the architecture to be as safe, secure, and private as possible."

However, the existence of safeguards does not mean all problems are solved.

If AI sends a wrong email, purchases a more expensive product than expected, or transmits sensitive information to a third party, who is responsible? New responsibility standards must be established among AI developers, platforms, users, sellers, and payment providers.

■ The Role of Digital Wallets in the AI Payment Era

For AI agents to make payments on behalf of people, more than just storing card numbers is required.

Can AI limit the amount it can spend?

Can purchases of specific product categories be blocked?

Can approval requests be sent to the user before payment?

Can incorrect transactions be reversed?

Can the identity of the AI that performed the transaction be verified?

Mastercard predicts that as agent commerce expands, financial institutions and sellers will need to newly manage AI agent identity, user consent, payment approval, and liability. In particular, a system that can verify the user's purchase intent and authority granted to AI even before payment will become important.

In this process, digital wallets and blockchain technology may also take on new roles. This is because each AI agent can be assigned a separate wallet or transaction authority, and transaction records and approval scopes can be made verifiable.

However, it is too early to conclude that blockchain and virtual assets will become indispensable for AI payments. Existing card and simple payment systems have strengths in terms of consumer protection and payment cancellation.

Currently, the spread of AI agents is more likely to be an opportunity to redesign authority management, identity verification, transaction records, and liability rather than an immediate replacement of payment methods.

■ Lifestyle Changes to Expect First in Korea

In Korea, the use of AI agents may first appear in the following areas:

Travel

If you request, "Book a trip to Jeju Island within this weekend's budget," it compares transportation, accommodation, rental cars, and restaurants all at once.

Shopping

Products can be selected by reflecting not only price but also delivery date, review reliability, return conditions, and past purchase history.

Work Tasks

It can schedule meetings, gather relevant materials, and even draft follow-up emails for each attendee after the meeting.

Subscription and Household Expense Management

It can compare prices for telecommunication fees, OTT, insurance, and regular delivery services, and cancel or change unnecessary subscriptions.

For services to become widespread in Korea, integration with domestic platforms, payment services, personal information protection regulations, and identity verification systems is necessary. Especially in areas with significant liability issues like finance, healthcare, and law, it is highly likely that AI will continue to require final human approval rather than making independent decisions.