Meta Platforms is betting that making AI agents easy to use will get mainstream consumers to start adopting them in their daily lives.

The company on Tuesday announced a new personal AI agent that can buy goods online, respond to emails and complete other tasks users authorize it to do for them. The agent, called Muse, is accessible via a dedicated app. Users operate the agent by conversing with it like a typical chatbot and can choose their own names for it.

The new tool is Meta's latest attempt to monetize its massive AI investments, which are expected to top $130 billion this year. The AI agent is free to access for "most of what people need" and has two tiers of subscription plans for power users: one for $20 a month and another for $100 a month, Meta said.

Anthropic and OpenAI have quickly built multibillion-dollar businesses around agents capable of performing complex computer tasks, but most of that usage is driven by software coding or other business applications. Consumer adoption of the technology lags behind, offering Meta, Apple and other companies a potential lane for catching up.

"The fact that we've given such a powerful technology in a package that is going to be really digestible for the product is one of the things that will set us apart from lots of other agentic products," Meta's AI chief, Alexandr Wang, said in an interview.

The company is also heavily promoting the safety and security of its new agents at a moment when researchers have been sounding alarms about the potential for AI to escape the control of humans.

In the last few months, models from Anthropic, OpenAI and Meta have gone rogue and hacked other companies, including open-source AI collaboration firm Hugging Face. In a recent report, outside researchers found that OpenAI's rogue agents colluded to deceive testers about the cheating they used to pass a test.

"We definitely are quite concerned about the potential for rogue agents," Wang said of Meta's overall AI efforts. But with respect to the new agents, he said, the model that powers them hasn't displayed the same capabilities as those involved in prior hacks.

Of more immediate concern, he said, is the potential for an agent to accidentally leak personal information or delete important emails.

To that end, each agent runs on its own "dedicated computer in the cloud" and no one else's agent can reach it, Meta said. A separate agent running on the same machine acts as a check on the Muse agent and doesn't allow it to reach the internet unless approved.

"Muse can and will still make mistakes, but we expect they'll be much less frequent and cause much less damage due to the safety systems we've built in," Meta said in a blog post announcing the new agents.

Users can choose which apps to connect to their agent, including their email, calendar and fitness apps, and control the degree of access they grant it. On email, for instance, users can choose whether the agent can only read emails or whether it can also craft and respond to emails on the user's behalf.

Once connected, the agent can book flights, buy movie tickets and shop around for other items on the web. It asks approval before completing any purchase, Meta said.

In a promotional video, the AI agent flagged a flight that had dropped by $40 and asked if the user wanted the agent to book it. It also reminded the user of an upcoming tee time and told the user which hole was going to be the most difficult on the golf course.

Meta said people can change their settings or disconnect from the agent at any time and can opt out of their data being used as training data for Meta's upcoming AI models.