An early tester at OpenAI forgot to send an invoice to the publisher. The AI agent reminded him about the mistake, prepared the invoice and mailed it only after he gave permission to do so. He had not asked for a reminder at all. The little story sums up the what Dots are capable of as always on agents unveiled by OpenAI on September 29.
It is ChatGPT but with a sweet name. It isn’t. The dot is an individual agent with its own computer in the clouds, its own browser and access to the applications that you connect it with. It is based on GPT-6 Astra and connects with more than 4,000 different applications. It needs your feedback for correction so that it knows how a good job is done by you. And you can access that computer anytime and see the results yourself. It is basically a colleague who works from your desk all day.
You access your dot from ChatGPT on desktop, web, mobile, Slack, and Teams. Messaging via texting is on the horizon. You could dial in for a voice conversation. The context follows you across channels, and your dot can send the first message with updates, questions, or decisions needing your input.
When you are not present, it conducts what OpenAI refers to as proactive research. This process leverages read-only capabilities within your integrated applications; therefore, it cannot send messages, modify content, or control your web browsing. Researching is fine, but acting requires approval.
The examples of OpenAI are real. There is a bug found in Slack and there comes a dot to check it out. The developer receives completed pull requests with video demonstrations on how to fix it. The scientist dot repeats his analysis once there are more data available, changing graphs. The creator dot highlights clips of the film and takes notes of the new interview transcript.
The main reason why people buy it is control. With Custom Rules, users can allow, block or request approval for certain activities. The Activity View reveals all background activity. The auto-revision stage verifies any process that interacts with your account or shares information. There are activities that will never be delegated, like the change of your password. A monitoring mechanism may stop a dot when it detects something suspicious.
However, OpenAI acknowledges that dots may err, and that’s why it advises to revise all consequential activity. An agent who does things on your behalf makes mistakes quicker than you detect them.
Dots are the equivalent of an assistant in your pocket. The grander vision for dots is about hiring practices. Specialized dots receive their own identity, credentials and access to the company’s infrastructure before performing specialized tasks. OpenAI has conducted pilots within its organization on procurement, invoicing, email marketing, customer service and business contracting. Enterprise piloting will follow and OpenAI is partnering with Microsoft to incorporate specialist dots into Agent 365.
Rollout of Dots started among the ChatGPT Pro and Business Premium members in qualifying regions. Dots will be available to all workspaces including those for Edu and Healthcare in beta upon activation by the admin. The first dot is included in your subscription package free of charge with additional allowances for in-depth conversations with extended limitations during the first month. Conversations with your dot are not counted in terms of your usage limits but tasks initiated through it in Codex/ChatGPT Work are.