Back to News

AI · Sep 30, 2026

OpenAI dots and the rise of always-on AI agents

At its DevDay event on September 29, OpenAI introduced dots, a new kind of agent in ChatGPT. According to OpenAI, dots are always-on agents that work in the background on their own cloud computer and can connect to more than 4,000 apps through plugins. They launched for Pro and Business Premium subscribers.

OpenAI was not alone. A few days earlier, Microsoft announced Autopilot, a persistent agent for Microsoft 365. Always-on agents are quickly becoming a product category, and I expect every major AI provider to ship its own version.

What "always-on" actually means

Most people use AI the way they use a search engine. You ask, it answers, the conversation ends. An always-on agent is different. You give it a job, and it keeps doing that job without being asked again.

Instead of "summarize this email," it is "watch my inbox and flag anything from a customer that sounds unhappy." Instead of "draft a report," it is "every Friday, pull this week's numbers and send me a summary."

The agent having its own computer matters because it can run tasks for a long time, use software, and keep working when your laptop is closed.

Useful jobs for an always-on agent

For a small or mid-size business, the best early uses are the recurring tasks that someone does every week and nobody enjoys.

  • A morning brief of new leads, overdue invoices, and calendar conflicts
  • Monitoring a shared inbox and drafting replies for review
  • Tracking competitor pricing or new listings and reporting changes
  • Following up on open quotes after a set number of days
  • Compiling weekly numbers from a few systems into one summary

Notice that most of these are about watching, gathering, and drafting. Those are the safe places to start.

The risks are different, too

An AI you talk to occasionally is easy to supervise. An agent that works in the background, connected to thousands of apps, is not. A few things become important.

Access. Every app you connect is something the agent can read, and possibly change. Connect only what the job requires.

Approvals. Anything that sends a message to a customer, moves money, or deletes data should wait for a person to approve. Most of these products now support approval steps. Turn them on.

Visibility. You need a log of what the agent did and when. If you cannot see it, you cannot trust it.

Cost. Background agents use compute while they work. Check how usage is billed and set limits.

Do not automate a mess

The biggest mistake I expect businesses to make with always-on agents is pointing them at processes that are not clear to begin with. If your team cannot explain how leads should be followed up, an agent will not figure it out either. It will just do the unclear thing faster.

Think first, AI second. Write down the job in a few sentences. If you cannot, fix the process before you hand it off.

What this means for your business

  • Always-on agents are now available from every major AI provider. Expect them to show up in tools you already pay for.
  • Start with watching, gathering, and drafting tasks, not tasks that act on customers or money.
  • Limit connected apps, require approvals for sensitive actions, and keep a log.
  • Define the job clearly before you delegate it.

If you want help picking the first recurring task to hand off, and setting it up with the right limits, we would be glad to talk it through.

Sources