OpenAI Dots: Full Setup and Live Demo

Summary

The video demonstrates OpenAI Dots, a proactive agent in the ChatGPT desktop app. The presenter describes its core idea as giving an agent an “ongoing responsibility”—for example, monitoring calendars or Slack and bringing relevant updates to you, rather than waiting for a prompt each time. She sees it as an accessible way for people who do not build agents with Codex or code to try proactive AI.

Setup and key features

• Desktop app: The presenter recommends downloading the updated ChatGPT desktop app and setting up a Dot there.
• Privacy setting: She advises personal-account users to turn off “Improve the model for everyone” under ChatGPT’s Data Controls if they do not want their input used to improve models. She notes that team and enterprise accounts handle this at the admin level.
• Virtual computer: A Dot can work in its own virtual computer, allowing it to browse and perform tasks without taking over the user’s local workspace. The presenter also enables access to her local computer.
• Customization: Users can name and customize their Dot, though the presenter encounters an error when trying to save an avatar.
• Natural-language instructions: The agent can be given tasks conversationally, including recurring or scheduled responsibilities. The presenter says its replies are fast and concise, with limited status updates rather than a detailed reasoning trace.
• Ways to interact: The demo includes chatting in the desktop app, adding the Dot to Slack, and attempting to call it by phone. The Slack setup works, but the call fails.
• Virtual-computer supervision: Users can watch the agent’s activity and take over its virtual screen to enter information. The presenter notes that, in her comparison, Muse offered a way to stop a virtual-machine action that she did not find in Dots.

Live demo

The presenter tests Dots with several personal and practical tasks:

1. Hair appointments: She asks it to help her remember haircuts and hair-color appointments, sharing her salon and scheduling habits. The Dot asks follow-up questions and offers to nudge her.
2. Events and tickets: She asks about Harry Styles and other Madison Square Garden concerts, and asks it to get ahead of ticket availability for a Woolworth Building tour.
3. Phone calls: When asked to call her salon, the Dot says it cannot make outbound calls. The presenter challenges this, but does not establish that it can place the call.
4. Hiring and purchases: She asks whether it can hire someone through Upwork and searches for Stevie Wonder tickets, including asking it to find a cheaper option. The demo also raises payment handling: the presenter says not to send card details in chat and discusses Link as a possible payment method, while observing that it was not integrated into the demonstrated flow.
5. Slack: She connects the Dot to a Slack workspace and gives it a recurring task to summarize useful AI use cases shared by her team or surfaced in alerts.

Analysis and verdict

The presenter frames Dots as part of a shift from prompting for individual actions toward setting goals, standards, or “north stars” and letting agents work toward them with available tools. She suggests that many specialized agents may eventually be coordinated by a higher-level agent.

Her early review is mixed: she likes the concept, natural-language setup, concise responses, and separate virtual computer, but reports bugs and missing capabilities. She considers Meta Muse and Instinct more capable at the time of filming, while expecting Dots to improve through updates. She recommends trying Dots as an introduction to proactive agents, rather than rushing to switch if already using more mature alternatives.

Access mentioned

At the time of the video, the presenter says Dots is available to Pro and Business Premium users and not yet available in the EU or UK. Availability may have changed since recording.

Main speaker and sources

• Main speaker: The video’s presenter, identified by the channel metadata as Allie K. Miller.
• Other sources discussed: OpenAI’s product demonstrations and statements attributed to Sam Altman; the presenter’s comparisons with Meta Muse, Instinct, Codex, Claude Code, and Slackbot.