OpenAI Dots Use Cases: What Teams Actually Use Them For

A general-purpose agent is hard to evaluate in the abstract. What helps is seeing a real workflow: the kind of work, the hand-offs, and the point where a person still has to decide. When OpenAI announced Dots on 29 September 2026, it described five such workflows rather than listing features. This page collects them and points to the deeper, role-by-role guides where the details matter.

All five share the same underlying abilities. Every Dot is powered by GPT-6 Astra, has its own cloud computer and a browser, can write and test code, connects to 4,000+ apps, learns from feedback over time, works 24/7, and can take on a project while handling several others. What changes between scenarios is the domain, the people who review the work, and the guardrails that make sense.

The five scenarios OpenAI highlighted

1. Developers: feedback into tested pull requests

A Dot watches recurring customer feedback, scopes small improvements and bug fixes, builds and tests them, and brings back complete pull requests with videos. The interesting part is not that code can be written, but that the loop — listen, scope, build, test, hand over — keeps running without someone relaying it by hand. Our Dots for developers guide covers the end-to-end flow, the review boundary, and how to split work safely.

2. Launch leads: materials that track a moving scope

Launch scope changes constantly, and every change ripples through documentation, messaging and internal briefs. OpenAI's launch-lead example describes a Dot revising launch materials as scope changes, so the artifacts stay consistent with the latest plan. There is no dedicated page for this role yet, but the mechanics are the same as the other use cases: the Dot drafts and updates; you approve the positioning that goes public. The explainer is the best starting point.

3. Sales: proposals that keep up with requirements

A sales lead's Dot checks customer requirements and account history against product documentation, finds what needs testing, builds a proof of concept, suggests a solutions engineer, and updates the proposal and test plan as things change. Commitments are the sensitive surface here, so approvals matter. See Dots for sales teams.

4. Content creators: transcript to published drafts

A creator's Dot learns the creator's voice, turns an interview transcript into clips, show notes and social posts for approval, carries edits across the materials, and learns what resonates. The human keeps the editorial call. See Dots for content creators.

5. Scientists: analysis that updates with evidence

A scientist's Dot learns the research question and how evidence is evaluated, reruns analyses as data arrives, investigates unexpected results, updates figures, and revises explanations while flagging what needs review. Scientific judgement stays human. See Dots for scientists.

What the scenarios have in common

Each example describes ongoing responsibility rather than a one-off request. The Dot holds context, keeps working in the background, and reports what it did through an activity view. Each also keeps a person at the decision that carries risk: merging code, approving a commitment, publishing in a voice, interpreting a result. Auto-review checks actions against your instructions, Custom Rules and safety requirements before consequential steps, proactive research uses read-only tools, and some sensitive tasks always stay with you.

Dots can make mistakes. Treat the scenarios as a map of where an always-on agent can absorb recurring effort — not as a claim that any of this runs unattended. The security and privacy guide covers the controls, and our free Agent Readiness Scorecard helps you decide which of your own workflows fit next.

Choosing your first scenario

The five scenarios are not equally easy to adopt, and the order in which a team gets value from them depends on how much of the work is already well defined. A developer loop with a fast test signal and a clear reviewer is relatively easy to start, because success and failure are visible. Content repackaging is a natural early candidate because the approval step is low-risk and the feedback is immediate. Sales and launch work sit closer to commitments, so they benefit from stricter rules and explicit sign-off. Scientific analysis demands the strongest reproducibility and review discipline of all. Whichever you choose, the pattern is the same: start narrow, keep a human at the decision, and widen only once the loop is reliable.

Source: OpenAI, Introducing dots.