Chris Curl,
Editor
Oct. 15, 2026
The next addition to your office will not need a desk.
But it will need access to your software.
OpenAI calls these new workers “Dots”: always-on AI agents designed to carry out tasks, follow projects, and keep working between conversations. Powered by Astra, they can operate across connected applications and communicate through tools such as Slack and Microsoft Teams.
The name sounds harmless. Almost cute.
The business implications are anything but.
For the last few years, most people have used AI as a better search box or a writing assistant. Ask a question. Get an answer. Copy something into another application. You still do the work of moving the project along.
Dots are supposed to change that relationship. Instead of asking software for help with each step, you give it an objective and delegate parts of the process. OpenAI describes agents that can update a sales proposal as customer requirements change, research a problem, analyze data, prepare documents, and build a working demo for review.
Whether they can do all of that reliably is the question. But the direction should be obvious.
We are moving from software you operate to software that operates on your behalf.
The Office Is the First Test
Think about how much of a working day disappears into coordination.
Finding the latest document. Updating a spreadsheet. Checking whether someone replied. Moving information between systems. Turning a meeting into a list of tasks, then reminding people to finish them.
None of that is particularly glamorous. Much of it is necessary. And it is exactly the sort of work agentic AI is being built to handle.
A useful agent does not have to replace an entire employee to change the economics of a business. It only has to remove enough repetitive work that the same team can accomplish more.
For a small company, that could mean pursuing customers it previously lacked the capacity to serve. For a larger company, it could mean reducing administrative layers. For employees, it could mean more time spent making decisions, or management deciding it needs fewer people.
Both outcomes belong in the conversation.
I am bullish on the productivity potential. I am much less comfortable with the assumption that everyone will benefit equally.
The person who learns to supervise these systems may become considerably more effective. The person whose job consists mainly of moving information between applications may face a much harder adjustment.
Giving AI the Keys Changes the Risk
There is a reason the article about agents going rogue is important.
A chatbot that gives you a bad answer creates one kind of problem. An agent that acts on that answer inside a connected system creates another.
Reported incidents involving experimental agents include unauthorized website access and attempts to bypass security controls. OpenAI has distinguished those testing incidents from its commercial offerings, but they expose the challenge of giving increasingly capable software room to act.
That does not mean every Dot is a rogue employee waiting to happen. It means autonomy needs boundaries.
OpenAI says users can establish rules governing what Dots can do and when they must ask permission, with explicit consent required for sensitive actions such as changing passwords or permanently deleting data.
Those controls are more than a footnote. They are part of the very product.
Businesses will need to know what an agent did, which information it used, and who approved its actions. If reviewing every step takes longer than doing the job yourself, the economics fall apart. If nobody reviews anything, the risks become unacceptable.
The winners will have to solve both problems.
Productivity Is Not the Same as Profit
This also changes how I think about the AI trade.
More capable agents could make software substantially more valuable to customers. They could also put pressure on businesses charging people for tools that agents eventually operate, combine, or replace.
Which applications remain indispensable? Which become interchangeable? Who owns the customer relationship when an agent sits between the user and the software?
There is no reason to assume the answers favor every incumbent.
Then there is the infrastructure bill. Always-on agents doing useful work could generate much more computing activity than occasional chatbot conversations. But demand alone is not enough. The work has to be valuable enough to pay for the compute, supervision, and mistakes.
That is the distinction I keep hammering on in Digital Dispatch: a technology can transform the economy without making every company attached to it a good investment.
The promise is enormous. But the price still matters.
Follow the Work
Dots are worth watching because they make the next phase of AI easier to understand.
Forget the argument over whether a benchmark proves AGI. Watch whether a business can hand a system a real assignment and get back something useful without spending the afternoon fixing it.
That is where the productivity story becomes measurable.
And if these agents become dependable, the implications extend well beyond the office. The same ambition (systems that perceive, plan, and act) runs through the robotics thesis. Operating a browser is not the same as operating in a warehouse, but both point toward an economy in which software takes a more active role in getting work done.
That is why I am following this transition through Digital Dispatch.
The opportunity is not simply to buy whichever company has the most charming agent avatar. It is to understand where value goes as businesses reorganize around these tools: the intelligence providers, the infrastructure beneath them, and the companies that put them to productive use.
Click here to watch my Nvidia Atlas presentation for the broader investment thesis.
You do not have to love the name “Dots.”
But if software starts doing the work instead of merely helping with it, you will want to understand what that means for the companies (and the jobs) you depend on.
Keep coming back,
Chris Curl
Editor, Bizarro World