OpenAI introduced Dots at DevDay on 29 September as always-on agents that can keep working between conversations, use connected apps, run scheduled tasks and bring results back for review. Their persistence changes the decision a business has to make: the work needs a boundary that still holds after the initial instruction.
For a small business, the sensible first job is concrete: ask a dot to check a small set of approved systems each morning and return the new, overdue, missing or conflicting items that need attention. A person still decides what happens next.
The useful change is persistent attention
OpenAI describes Dots as agents with their own cloud computer that can use connected apps, continue ongoing work and report progress or decisions that need a person. Users can also schedule recurring checks and set rules for when a dot may act, must ask or should hand work back.
That could matter to a team that repeatedly checks several systems to answer the same questions: What arrived overnight? What is overdue? Which record is missing information? What needs attention before today's meetings?
Begin with the work of noticing and organizing what needs attention. Keep the authority to act with the person responsible for the workflow.
Start with one recurring check
Choose a recurring monitoring task that one person already understands. A useful check might run each weekday morning and report:
- new items that match a written definition;
- overdue work and the person currently responsible;
- missing or conflicting information;
- upcoming deadlines or meetings that need preparation; and
- suggested next actions, each linked to its source.
Use only the connected sources required for the check. OpenAI says a dot's proactive research uses read-only tools, while other actions are governed by permissions, custom rules and automatic review. In practical terms, the dot may inspect information and draft a response, but it should not send the response, change a record, make a purchase or commit the business during this first test.
| Test boundary | First-week setting |
|---|---|
| Scope | One named monitoring workflow |
| Inputs | The minimum approved sources |
| Output | One sourced attention list at a fixed time |
| Actions | Check, summarize and draft; no sending or record changes |
| Review | One named person checks every result |
| Escalation | Flag uncertainty instead of guessing |
| Duration | Seven working days |
If you cannot name the workflow owner, the intended reader of the result and the conditions that deserve attention, the workflow is not ready for the agent.
Measure the review burden as well as the output
During the seven-day test, record five things:
- important items the check found;
- important items it missed;
- items it flagged unnecessarily;
- facts or recommendations the reviewer corrected; and
- the time required to review the result.
Those measures show whether the agent is making the work easier to supervise or merely creating another inbox.
OpenAI's own launch guidance recommends starting with one recurring task, approved sources, a named owner and reviewer, and a result someone can check. The company also says Dots can make mistakes and that consequential work should be reviewed.
There is a further reason to keep the boundary simple. In one challenging evaluation of chained tasks, OpenAI reported flag rates for moderate-severity scope violations of 8.6% with five intervening tasks and 19.7% with ten. OpenAI says these selected tests are not necessarily representative of production, so those figures are not a predicted failure rate for your business. They do show why broad instructions and shifting responsibilities make a poor first test.
Make the Build, Buy or Skip decision after the test
Treat Dots as a Buy when the product can reliably surface the right items from the systems you already use and the review effort is reasonable. You may not need a custom build.
Consider Build only when the test exposes a valuable gap: a required source is unavailable, the business needs a defined transformation the product cannot perform, or the handoff to an existing system requires a controlled connection.
Choose Skip, at least for now, when the workflow changes every day, source records are unreliable, nobody owns review, or the cost of a wrong action is too high for the available controls.
Who should wait
Wait if access has not reached your account, the necessary connected apps are unavailable, or your data and permission requirements have not been reviewed. Teams handling sensitive, regulated, financial, employment or customer-facing decisions should establish their policies and review boundaries before connecting those workflows.
The test is also unnecessary when checking one reliable source already takes only a few minutes and missed items carry little consequence. Adding an agent would create another system to supervise without solving a material problem.
Availability also varies by plan and market. OpenAI says Dots are rolling out gradually to Pro and Business Premium users in eligible markets, with an admin-enabled beta for Enterprise workspaces.
The product may eventually earn more authority. Let the first seven days earn the next permission.
Source note: This Field Note draws on OpenAI's launch documentation, controls guidance, rollout material and safety evaluation. We will update it as product behaviour and availability change.
Start by mapping the workflow, its exceptions and its owner. If you are deciding among several candidate processes, an AI Opportunity Review can compare up to three and define the first justified test.