AI assistants are starting to outgrow the chat window.
During the OpenAI DevDay on Sept. 29, the company introduced Dots, persistent AI agents designed to work across connected applications and continue tasks after users step away. The launch puts OpenAI on a similar path to Meta, whose Muse agent can also pursue goals, use online services, and act with less step-by-step direction.
But the companies emphasize different parts of the personal AI experience. OpenAI is positioning Dots heavily around ongoing knowledge work, while Meta is showing Muse handling a broader mix of personal errands, purchases, travel, and planning.
The bigger shift is the same: personal AI is moving from answering requests toward taking responsibility for tasks that unfold over time.
OpenAI is building Dots around ongoing work
OpenAI’s Dots run on GPT-6 Astra and receive their own cloud computer and browser.
They can connect to more than 4,000 applications through OpenAI’s plugin ecosystem and communicate with users through ChatGPT, Slack, and Microsoft Teams. OpenAI also plans to add text messaging.
Rather than requiring a new prompt for every step, a Dot can be assigned an ongoing objective and handle portions of the work independently.
OpenAI gives the example of a software-development Dot monitoring customer feedback, finding recurring problems, creating and testing fixes, and returning pull requests for review. Another Dot could update scientific analyses as new data arrives, while a sales-focused Dot could revise proposals and proofs of concept as customer requirements change.
Those examples place Dots firmly inside workflows that would normally require repeated human follow-up.
OpenAI is also previewing specialist Dots with separate identities for access management and deeper connections to company systems. Longer term, the company says multiple Dots could work together.
For businesses, that creates a different management problem from deploying a conventional chatbot. IT teams may eventually need to decide not only what information an AI can access, but which systems it can operate and which actions require human approval.
Meta’s Muse reaches further into everyday life
Meta’s Muse has a broadly similar technical foundation.
It operates through a dedicated cloud-based virtual machine with its own browser and can continue tasks after a user closes the app. People can access Muse through Meta’s dedicated application or WhatsApp.
Meta’s examples, however, extend well beyond workplace tasks.
The company has shown Muse helping arrange travel, send emails, organize events, negotiate bills, create shopping lists, make purchases, and adjust longer-term plans as circumstances change.
Muse also supports payments through Stripe’s Link using one-time card numbers, allowing transactions without exposing a user’s primary card number directly to the agent or merchant. Meta says support for Shop Pay and 1Password is planned.
The difference between the two products is therefore less about what they theoretically can do and more about where their creators are placing the emphasis.
Muse could also handle work, and Dots can perform personal tasks too, but OpenAI repeatedly demonstrates Dots inside work processes. Meanwhile, Meta presents Muse as an agent that can follow users across more of their everyday digital lives.
Meta also plans to bring Muse to its AI glasses, potentially extending the agent beyond traditional phone and computer interfaces.
The real change is delegation, not better prompting
The most important similarity between Dots and Muse is not their browsers, cloud computers, or integrations. It is the amount of responsibility users are being encouraged to hand over.
Most generative AI tools still operate request by request. A user asks for something, receives an answer, and returns with another prompt when more work is needed.
Persistent agents are designed to reduce that back-and-forth.
A Dot can follow an ongoing assignment and return when work has progressed or a decision is required. OpenAI also supports recurring tasks and limited proactive research across connected information.
Muse follows a similar model, continuing tasks when the user is elsewhere and returning when it reaches a step that requires approval.
For workers, that changes the skill being asked of them. Instead of only learning how to write better prompts, they may increasingly need to define outcomes, set boundaries, review results, and decide where human judgment is still necessary.
More autonomy requires more access
The trade-off is straightforward: an AI cannot arrange travel, update a project, browse business systems, or make purchases unless it can access the information and tools needed to do those things.
Both OpenAI and Meta have therefore built approval and monitoring controls around their agents.
OpenAI lets users decide which applications a Dot can access and create rules that allow, block, or require confirmation for particular actions. OpenAI says some sensitive activities, including changing a password, remain reserved for the user.
Users can also inspect what a Dot is doing through its cloud computer.
The company also places additional restrictions on proactive background research. During that process, Dots can use read-only tools but cannot send messages, modify connected applications, or take control of the user’s browser or computer.
For Meta, their approach is different. Its Sentinel system monitors Muse from within the agent’s secure virtual environment and reviews its internet activity. Muse also asks for confirmation before certain consequential actions, including sending an email or completing a purchase.
The company says stored credentials are handled so Muse cannot directly view them, and users can review a record of completed and planned actions.
Those safeguards matter because persistent agents potentially receive much broader access than traditional chatbots. OpenAI also cautions that Dots can still make mistakes and recommends human review for consequential work.
For businesses and individual users alike, the question is no longer simply whether an AI produces a good answer. It is how much authority it should receive before a person needs to step back in.
More must-read AI coverage
- SS&C Intralinks DealCentre AI vs. Datasite: Which platform is built for the future of dealmaking?
- SS&C Intralinks FundCentre AI vs. Juniper Square: Which platform better supports modern private markets fund managers?
- Why Data, Not Models, Determines AI Success
- The Rise of the AI-Native Factory: How Physical AI Is Transforming Manufacturing
Your next AI assistant may be something you supervise
It is too early to know how much autonomy users will ultimately give persistent AI agents.
But OpenAI and Meta appear to agree on the direction of travel.
The next personal AI may remember an ongoing objective, watch for changes, complete intermediate steps, and return only when it needs a decision.
That changes the user’s role from constant operator to supervisor.
Instead of focusing only on prompts, users may need to decide what an agent can access, which responsibilities it can handle independently, what requires approval, and when a person should take control again.
OpenAI is starting with an agent embedded deeply in ongoing work. Meta is reaching more broadly into everyday digital life.
For now, the more important question may be less about which assistant appears smarter and more about which one offers enough control, visibility, and reliability to earn responsibility for real work.
Want to learn more AI tips, tricks, and prompting techniques? TechRepublic readers get free 7-day access to The Neuron Academy, our practical learning platform designed to help professionals use AI more confidently at work. Browse all lessons →
Let us teach you How to Talk to AI for free! Try our six-minute course at The Neuron Academy and learn a few simple ways to write better prompts and get more useful results from AI, or browse our other AI course for free for seven days. Check out all the lessons here →