Your newest coworker never clocks out
OpenAI just launched an employee that never sleeps. It never takes a lunch break, either. That sounds great until you ask who is supervising it.
The product is called Dots. Your team might already be able to use one. So let’s talk about what it is. Then we will cover what you should do about it before Friday.
What OpenAI actually announced
On September 29, 2026, OpenAI held its DevDay event. Among more than 20 announcements, one stood out for business leaders. OpenAI describes Dots as “always-on agents built to handle everything.” An agent is software that takes actions for you, not just answers questions. Always-on means it keeps working after you close the chat window.
Engadget’s live coverage reported that Pro and Business Premium subscribers get one free Dot. They can pay for more. In one demo, a Dot took forwarded Slack threads and worked on them. Sam Altman also described long tasks, such as moving an app off an API that is about to be retired.
The rollout looks different for larger organizations. OpenAI says Enterprise, Edu and Healthcare workspaces can try the beta only when an admin turns it on. In its words, it is “off by default.” That is a good choice, and it deserves credit.
OpenAI also announced a privacy layer it calls Private Intelligence. It includes Zero Data Retention with Private Safety Processing. In plain English, safety checks can run without OpenAI staff seeing your content. A second piece, Private Inference, is only a preview and is listed as “coming this fall.” Therefore, do not count on it yet.
The detail most coverage is missing
Most headlines focus on the shiny demos. However, the more useful story sits in the fine print about who holds the switch.
The off-by-default rule applies to Enterprise, Edu and Healthcare workspaces. The recap page does not say the same for Pro. That matters because Pro is a personal plan. An employee can sign up, get a free Dot, and connect it to work tools with their own approval. Your admin never sees that step.
OpenAI also says users “choose which plugins to use and approve the access each one receives.” In other words, the person clicking Approve is the employee, not your security team. That is shadow AI with a job title. Shadow AI means AI tools your staff use without company approval.
The demos also showed real rough edges. Engadget reported that one Dot interrupted a presenter mid-talk. Voice chat failed on stage, too. Those are small glitches. Still, an agent that works while nobody watches needs better odds than a stage demo.
Why this matters to your organization
Think about the Keep Human bucket in my A.S.K. Framework. A.S.K. stands for Automate, Share, Keep Human. It sorts every task into one of those three groups. An always-on agent quietly moves tasks between them, and nobody decides that on purpose.
For example, a Dot that answers a customer thread is now making a judgment call. A Dot that edits a shared file is now changing your records. Because the agent never stops, mistakes also run longer before anyone notices. You saw this earlier this month. An OpenAI monitoring alert fired, but the run still continued for hours before a person shut it down.
So the question is not whether to ban Dots. Instead, ask who decides what a Dot may touch, and who reviews what it did. Meanwhile, your competitors are asking the same thing. The ones who answer first will move faster, because their people will trust the tools.
This is also a governance question. Governance means the rules, owners and reviews that keep AI useful and safe. Without it, each employee writes their own rules. As a result, you get ten different policies and no record of any of them.
What to do this week
- Check your admin settings first. If you run an Enterprise workspace, confirm that Dots are still off. Then write down who has the power to turn them on.
- Ask your team who already uses one. Do it as an open question, not an interrogation. Then you will learn more, because people answer honestly when they feel safe.
- Publish a one-page agent policy. List which tools an agent may connect to, and which it may never touch. Payroll, customer records and legal files belong on the second list.
- Name a human owner for every agent. One person should review its activity each week. If no one owns it, it does not run.
- Start with a low-risk pilot. Pick one task, such as summarizing meeting notes. After that, measure the time saved and the errors caught before you expand.
- Ask the vendor about logs. The recap page I read does not mention audit logs or admin policy controls. Get those answers in writing before you scale.
Let’s talk before you need to
Agents are coming into your company whether you plan for them or not. The organizations that win will be the ones that decide the rules first. If you want a second set of eyes on your agent policy, reach out to me at UnDesto AI. We can walk through it together in one conversation, and you will leave with a clear next step.
***
Want more real talk about AI at work?
Who’s supervising the AI coworker is exactly the kind of question we get into on our podcast, OMG AI! Listen on Spotify for honest, no-hype conversations about what AI means for leaders, teams, and the future of work.
🎧 Listen to OMG AI! on Spotify
Our new AI Horror Stories series is also live on YouTube. These are real cautionary tales about AI gone wrong and what you can learn from them before it happens to you.
▶️ Subscribe to Learn with Myra on YouTube
***
Sources
- DevDay 2026 Recap – OpenAI
- OpenAI Dev Day 2026: Live updates – Engadget
- OpenAI DevDay recap – CNBC (headline and date only; page could not be fetched in full)
Photo credits: control room by Frantisek Duris on Unsplash; team meeting by Lyubomyr Reverchuk on Unsplash.
CONTACT US TODAY!
