OpenAI launches Dots to take on Meta’s Muse with AI agents that can act on your behalf

OpenAI launched Dots on Tuesday, its new “always-on” AI agents designed to work toward users' goals across apps, at its DevDay event in San Francisco. The launch comes less than a month after Meta introduced Muse, its personal AI agent that can handle tasks such as sending emails, booking travel and shopping.

The two products share several similarities. Both have their own cloud-based computing environments, can work autonomously for extended periods, remember information about users and can take actions across applications.

But the companies are approaching the technology from somewhat different directions. OpenAI is pitching Dots heavily as an AI teammate for work, while Meta has positioned Muse as a personal AI agent for everyday tasks.

OpenAI describes Dots as agents that can take a project and “run with it” rather than requiring users to direct every individual step.

Each ‘Dot’ has its own cloud computer and browser. Users can connect more than 4,000 apps through OpenAI's plugin ecosystem, while the agents can also be accessed through ChatGPT, Slack and Microsoft Teams.

Also Read: OpenAI bets on always-on AI agents as safety concerns test pace of AI development

For example, OpenAI says a developer could have a Dot monitor customer feedback, identify recurring bugs, build and test fixes, and return complete pull requests for review.

The company also sees Dots handling work that changes over time. A sales Dot could track changing customer requirements, update a proposal and test plan, and flag issues that require human attention.

OpenAI is initially rolling Dots out to Pro and Business Premium users in eligible markets. Enterprise users can access a beta when their workspace administrator enables it.

The company is also developing specialist Dots for organisations. These would have their own identities, credentials and access to company systems, allowing them to take on specific responsibilities such as procurement, invoice processing, customer support and commercial contracting.

What is Meta Muse?

Meta introduced Muse on September 8 as a personal AI agent designed to proactively work towards a user's goals.

Users can give it a broader goal, after which it can develop a plan and continue working after the user closes the app. It can return when something changes or when it needs approval.

Also Read: Meta expands Muse AI agent for small businesses

Muse also has access to information a user has previously shared. Meta says this allows it to make suggestions without being prompted each time, such as turning a recipe saved on Instagram into a grocery list or remembering dietary restrictions when planning a dinner.

Both want AI that keeps working when you are not

The biggest similarity between Dots and Muse is the move from prompt-and-response AI to background AI.

With a chatbot, a user generally asks a question and waits for an answer. With an agent, the user can give the system a goal and let it work through multiple steps.

OpenAI's examples include a Dot updating research as new data arrives, revising a sales proposal as requirements change or turning customer feedback into tested software fixes.

Meta has similarly described Muse adjusting a training plan as a person's circumstances change, negotiating a sale or continuing a task after the user has closed the app.

That is also why both companies have built dedicated computing environments for their agents.

An agent that can browse websites, access applications and perform actions needs somewhere to operate. OpenAI says a Dot's computer is separate from the user's computer unless the user explicitly connects it. Meta's Muse Secure VM similarly separates the agent from the rest of the user's computing environment.

What happens when agents get it wrong?

The promise of autonomous AI comes with a corresponding risk. The more an agent can do without asking, the more damage it could potentially cause when it misunderstands an instruction.

That concern has already surfaced around Muse.

Earlier this week, tech YouTuber Matt Robb said Muse shared his home address with a prospective Facebook Marketplace buyer after he gave the agent hands-off control of his Marketplace account. According to Robb, the agent also agreed to a lower price without informing him until later.

Robb had provided Muse with his address and pickup details, but said he did not expect the agent to share the address with every buyer. He also said he had selected an “Allow Always” permission option without realising it would allow Muse to send messages on his behalf without further approvals.

Also Read: Meta bolsters Muse safety warning after security vulnerability found

The incident came after Meta patched a zero-day vulnerability that could have allowed local attackers to take control of Muse. Amazon has also reportedly blocked Muse from accessing its retail platform over concerns about customer credentials.

The liability question is getting harder

The issue is no longer just whether an agent can make a mistake. It is increasingly about who bears responsibility when an autonomous system causes harm.

Anthropic, which is also developing highly autonomous AI agents, flagged the issue in its prospectus for its planned stock-market debut, according to Reuters.

The company said that it is unclear how existing laws would apply to AI agents, including whether an agent's actions would legally bind the user who deployed it, and whether AI systems would be treated as products, services or something else.