OpenAI Dots are office agents. Treat them like interns.
Always-on work agents are not just chatbots with avatars. The real question is what you let them do without you watching.
Short answerTry Dots only for narrow, reviewable work loops first; do not connect every app on day one.
By JasonPublished Sep 30, 2026Last verified Sep 30, 20265 min read
OpenAI's Dots are being pitched as always-on workplace agents: named assistants that can sit inside ChatGPT, Slack, Teams, and OpenAI's work environment, keep track of assigned tasks, and report back when something changes. That sounds useful if you are a founder, developer, operator, or content team already drowning in small follow-ups. It also changes the risk profile. A normal chatbot waits for a prompt. A Dot is described as proactive, persistent, connected to work apps, and capable of taking on complex work or coding tasks. That makes the buying question less about whether the demo looks clever and more about delegation design: what should an agent be allowed to read, what should it be allowed to change, and where should human approval be mandatory? The early coverage points to a product that is interesting for teams already paying for upper-tier ChatGPT plans, but too consequential to treat like a cute productivity toy.
What Dots are supposed to be
OpenAI describes Dots as proactive assistants that can keep working across complex projects and everyday tasks. PCWorld frames them as workplace delegates: named agents you can chat with through ChatGPT, text, phone, Slack, Microsoft Teams, or ChatGPT Spaces. Lifehacker adds the more operational detail: users create an individual dot for a specific purpose, the agent learns from responses over time, and OpenAI says it can connect to more than 4,000 apps at launch.
That is a different promise from a chatbot. A chatbot answers. A Dot is described as watching, continuing, reporting back, and sometimes acting before you ask. Lifehacker's examples include identifying a bug and deploying a fix, preparing an invoice for approval, monitoring customer feedback, and drafting social posts from interview transcripts. Those are not search queries. They are work loops.
The useful version is narrow
A strong first use case is not "run my company." It is a boring, bounded loop with a clear stop condition: watch a support channel and draft a triage note, scan customer feedback and propose small product fixes, turn call transcripts into a shortlist of clips, or prepare a weekly issue digest for the engineering channel.
Those jobs have three things in common. The inputs are known. The output can be reviewed quickly. A bad answer is annoying, not catastrophic. That is where an always-on agent can be useful without becoming an invisible manager of your business.
Who should care now
Lifehacker says launch access is limited to paying users including ChatGPT Pro, Business Premium, and Enterprise users, with one dot available at launch and more promised later. PCWorld also describes Dots as requiring a Pro plan or better, with one included per eligible account.
That makes this early version most relevant to people already paying for high-end ChatGPT access and already living in Slack, Teams, or coding workflows. If you are not in that group, the right move is to watch how teams build permission templates, review logs, and approval flows around Dots before copying the workflow.
The interesting part of Dots is not the avatar or the name. It is the permission boundary. If the product works as described by OpenAI and the early write-ups, a Dot is closer to a junior operator with app access than to a tab in ChatGPT. That is why small teams should evaluate it backwards: start from the actions you would be comfortable undoing. Monitoring feedback, summarizing transcripts, drafting invoices, and preparing pull-request notes are plausible first jobs. Deploying fixes, sending client email, or touching billing systems should sit behind explicit approval until the audit trail and failure modes are boring. OpenAI is emphasizing control, separate cloud computers, permission gates, and auto-review, which is the right vocabulary. But vocabulary is not operations. The teams that benefit first will be the ones with clean scopes, written rules, and existing review habits. Everyone else should resist the urge to connect every app on day one.
Try Dots only for narrow, reviewable work loops first; do not connect every app on day one.
Dots look most promising as narrow workplace delegates, not general-purpose digital employees. Start with one low-risk loop, write rules for what the Dot may read and change, and keep approval gates around customer-facing, financial, and production actions. If you are not already paying for an eligible plan or using work apps heavily, there is no reason to rush.
People who only use ChatGPT for one-off answers, teams without permission hygiene, and anyone expecting a fully autonomous employee replacement.
Read next
Follow new articles
Email updates are not live yet, and we are not collecting addresses. To follow new articles, use the RSS feed.