Tech Companies Are Betting Big On AI Agents. Goldman Says Ads And Subscriptions Could Be The Next Step.

AI companies are moving beyond chatbots that answer questions and toward agents that can carry out tasks for users, putting a new focus on how companies will make money from software that can shop, arrange travel, manage calendars and interact with other online services.

Goldman Sachs Research's Eric Sheridan said consumer AI agents represent a shift from conversational software toward systems that take action on a user's behalf. His comments followed Meta's Sept. 8 launch of Muse, its new personal AI agent.

"I think we are at the beginning of a paradigm shift. We are going from a conversational relationship with these agents to a much more action-oriented relationship," Sheridan, business unit leader of Goldman Sachs Research's technology, media and telecommunications group, said in a post published by the bank.

Goldman sees advertising and paid subscriptions as two ways companies can generate revenue from consumer AI agents. Sheridan said the mass market could resemble the existing web, where users can pay for premium services or use free products supported by advertising.

"The mass market for AI agents will be monetized with advertising and subscriptions, much like the way the web operates today," Sheridan said in the interview.

The bank also sees AI being used more extensively in the advertising business itself. Sheridan said the creation, placement and measurement of digital ads still involve substantial human input, while AI and machine-learning systems are increasingly being used across those processes.

Meta launched Muse in the U.S. on Sept. 8, allowing users to connect the assistant to services including email, calendars, payments, shopping and smart-home applications. It can send emails, book travel and perform other tasks through connected services.

Muse is available through a standalone app and WhatsApp, with Meta offering a free tier alongside paid subscription plans. The company launched $20-a-month and $100-a-month plans for users who want higher usage limits.

Muse overtook ChatGPT as the leading free iPhone app in the U.S. last week and recorded about 730,000 downloads over roughly five days following its Sept. 8 release, according to Sensor Tower data.

Meta shares surged Monday after Muse climbed the App Store rankings. The company is also encouraging outside developers and businesses to create connectors that allow the assistant to interact with their services. Stocks extended gains on Tuesday.

Other major technology companies have been developing similar tools, with OpenAI, Anthropic and Google introducing agents that can interact with websites and software and perform multi-step tasks. Retailers and other businesses have also begun testing agents designed specifically for online shopping.

Anthropic earlier this month released blueprints that retailers can use to build shopping and merchant agents using Claude. The tools can make product recommendations and add items to shopping carts, while Anthropic said one partner using its technology recorded cart sizes about 30% to 35% larger and customers were about 60% more likely to complete a purchase.

Banks have also begun examining how AI agents handle transactions and customer information. A group that includes Bank of America, Capital One, ING, NatWest and Commonwealth Bank of Australia warned Tuesday that AI shopping agents present risks involving fraud, scams, incorrect purchases and data privacy.

The banks called for discussions with regulators and technology companies over safeguards for agent-based commerce, including disclosures when AI is involved in transactions and stronger protections for customer data and payment information.

Meta says users decide which applications Muse can access and can revoke that access. The company has also said account credentials are kept in secure storage rather than being visible to Muse itself.

Goldman's Sheridan said access to calendars, passwords and payment credentials increases the number of tasks an agent can perform. He also identified consumer trust as an important factor as agents gain access to more personal information and online accounts.

Open-source alternatives are developing alongside products from major technology companies, with OpenClaw, a framework created by Austrian developer Peter Steinberger, attracting developers earlier this year. The software gained particular traction in China, where companies and programmers have used it to build AI agents.