An agent that keeps working can save time, but every background action needs a bounded scope, visible log and a human stop point before it can affect the outside world.
Confirmed facts
OpenAI introduced Dots on Sept. 29, 2026, as part of a broader set of agent and developer announcements around DevDay.
The product is described as supporting ongoing or scheduled work instead of requiring a fresh, one-shot prompt for every step.
The announcement and recap describe product plans and availability, while exact access, account permissions and workspace controls depend on the configured plan and rollout.
A product announcement does not by itself prove that an agent can safely complete a real task without supervision or that every integration is available to every user.
Signal Wire analysis
The shift is from “ask and receive” to “delegate and review.” That makes logs, handoffs, retries and cancellation part of the product, not optional polish.
Background work also creates a bigger blast radius: a stale instruction can trigger a message, change a document or spend a budget after the user has left the screen.
The right comparison is not only model quality. Measure how clearly a user can scope a task, inspect intermediate steps, stop it and recover from a wrong action.
For an early trial, keep Dots on read-only sources or reversible drafts, and require approval for money, deletion, account changes and public posting.
What is still unknown
- Which plans, regions and integrations support Dots at each stage of the rollout.
- How logs, data retention, retries and user cancellation work in long-running tasks.
- What limits or liability rules apply when an agent acts through a connected account.
How readers can check
- Start with a read-only task and set a clear time, data and spending boundary.
- Review the activity log and permissions before connecting email, payments, code deployment or social accounts.
- Keep the final send, publish, purchase or delete action behind a human confirmation.
What to watch next
- Dots availability and permission documentation
- Audit-log, retention and cancellation controls
- Independent reports of long-running agent failures and recovery
This article provides information and research context, not individual investment, legal or financial advice. Dates and policies can change; check the latest original source before making a decision.