OpenAI Dots: Inside the $100 Always-On Agent

The always-on AI personal assistant has arrived as the tech world’s latest bet on consumer software. These tools work in the background on your computer even when you are not typing prompts.

Startup Instinct gained quick attention this year with its smart agent. Meta followed by putting heavy marketing behind its assistant, Muse. Now, OpenAI is pushing into the field with Dots, an agent avatar released in late September.

Alexander Embiricos leads OpenAI’s enterprise team. At TechCrunch Disrupt, he explained how Dots works and where the software is headed. He described why AI firms use mascot avatars. He also addressed privacy concerns and revealed talks with payment networks.

Factsheet: Always-On AI Assistant Landscape

The Shift From Prompting to Background Work

For two years, generative AI relied on a prompt box. A user typed a question, waited for words to appear, and copied the result.

Persistent agents remove that manual work. Instead of waiting for commands, an always-on assistant watches your calendar, email inbox, and open apps.

Embiricos noted that users could already build similar setups with OpenAI Codex. Yet running workflows in Codex required manual coding and close supervision.

“On the product team, what I think about is, how do we make this intuitive and useful to people?” Embiricos said at Disrupt. He stressed that success depends on how the tool feels to real users.

Embiricos said his wife leaves her computer on all night. Her agents keep working while she sleeps. This ongoing work drives token usage for OpenAI. For users, it provides help that plans ahead for upcoming tasks.

The Avatar Metaphor: Why AI Giants Choose Friendly Mascots

One clear trend across new AI assistants is the move toward friendly mascot avatars. Instead of a plain text box, OpenAI created a bubbly avatar named Dots. The name comes from the white space in the OpenAI logo.

“There is kind of this convergent evolution happening,” Embiricos said. He noted that framing the software as a single ongoing conversation makes interacting with complex models feel natural.

Treating the assistant as a persistent entity helps users build mental models of what the software knows. Users do not have to start fresh chat threads. The agent remembers past context across weeks of work.

The approach mirrors efforts across the industry. Meta built Muse as a lively companion. Startups like Instinct use desktop widgets to stay active on screen.

Proactive Research and Autonomous Problem Solving

The real value of an always-on assistant lies in proactive research. Dots does not simply wait for a query. It regularly reviews recent chats, scans calendar entries, and drafts proposals on its own.

“On a cadence, it will think about what it knows about you,” Embiricos said. “It reviews past chats and looks at the tools you shared with it.” The agent then conducts web searches and drafts solutions before the user asks for help.

During early testing, timing remains a challenge. Embiricos recalled his own agent alerting him to a meeting conflict only three minutes before the event started.

Yet when applied to technical infrastructure, background monitoring has shown promise. Embiricos cited an engineer at a drone delivery startup whose system suffered a server outage. The engineer logged on to check the alert. By then, their Dot had already found the bug, pulled logs, and proposed a fix.

The Privacy Dilemma: Granting AI Access to Personal Data

To be genuinely useful, a personal assistant requires deep access to private communications. Dots connects to calendars, email accounts, desktop files, and cloud storage drives.

This access creates an unavoidable tension between utility and user privacy. Meta has faced criticism for using Muse to build detailed profiles of user social circles, which can feed its digital advertising engine.

OpenAI maintains that Dots does not serve advertisements. Users can disable model training to prevent OpenAI from using their personal data to train future models. This aligns with broader enterprise moves, such as how Cornell University secured strict zero-training privacy guarantees for campus accounts.

Still, permission controls remain coarse. In its initial release, Dots often asks for wide-ranging permissions and refuses read-only access to cloud drives.

“What I think we need to do is basically build out super, super granular controls,” Embiricos said. He explained that OpenAI initially set conservative permission defaults, leading the agent to frequently ask for user confirmation before taking actions. The product team plans to refine these thresholds based on user feedback.

The Scaling Frontier: Payment Processor Links and Autonomous Commerce

As personal agents expand, the next operational hurdle is interacting with the external economy. While automated booking tools have grown popular, many human service workers and business owners dislike interacting with software bots.

Embiricos acknowledged this friction directly. Despite leading enterprise agent development at OpenAI, he maintains a human assistant for partner interactions. He added that he personally avoids attending meetings staffed by another person’s AI agent.

The biggest step forward will be giving agents the ability to spend real money. Allowing an assistant to book flights, reserve hotel rooms, or purchase hardware requires direct financial integration.

“We are deep in lots of conversations with payment processors,” Embiricos revealed.

Letting an automated model execute financial transactions would represent a major milestone for OpenAI. Yet giving an AI model access to credit cards or bank accounts will test consumer trust far beyond sharing an email inbox.

Frequently Asked Questions

What is OpenAI Dots, and how does it work?

OpenAI Dots is an always-on AI personal assistant integrated into ChatGPT desktop and mobile apps. It operates continuously in the background, analyzing schedules, emails, and files to proactively complete tasks without real-time prompts.

How much does OpenAI Dots cost?

Dots is currently available to subscribers on the ChatGPT Pro tier and higher enterprise tiers. Pricing for these plans starts at $100 per month.

How does OpenAI protect user privacy within Dots?

Dots operates under standard ChatGPT privacy policies. Users can turn off model training so that their inputs and files are not used to train future OpenAI models. OpenAI does not sell user data to advertisers.

Can OpenAI Dots make financial purchases for users?

Not yet. OpenAI is talking with major payment firms. The teams are testing secure ways for agents to carry out financial transactions on behalf of users.

In related enterprise security findings, read how Cohere flagged unexpected outbound upload behavior during security evaluations of Alibaba’s Qwen model.