Tech • AI • Robotics • Game
ChatGPT Dots Explained: Set Up One Useful AI Job (Safely)
6/10TL;DR
Dots in ChatGPT are designed to keep a narrow, ongoing task moving between conversations, but they work best when given clear sources, update rules, and approval limits with a human retaining final control.
KEY POINTS
A persistent AI task, not a one-off chat
A standard chat answers a request and stops, while a dot can continue tracking assigned work over time. It can operate from its own cloud-based environment, use designated context, and return when a decision is needed. That allows it to maintain continuity on a defined job such as monitoring a product launch.
Best use starts with one narrow responsibility
Rather than trying to run an entire workflow, dots are better suited to a single concrete assignment. In a product launch example, the goal is not a long recap but a short decision brief covering deadlines, unresolved questions, blockers, and what changed since the last update. Keeping the scope narrow reduces noise and makes oversight easier.
The core setup uses four instruction parts
A reliable assignment includes source, outcome, update trigger, and approval boundary. For example, the dot can be told to use a launch brief and a sanitized task list, track deadlines and decisions, and produce a concise brief focused on issues needing attention. Naming the source material directly is critical because vague missions lead to weaker results.
Monitoring must be explicitly defined
Connecting a messaging app or inbox does not mean the dot will automatically watch everything in it. Users need to specify what it should monitor and when it should report back. A useful rule is to notify only when a deadline is at risk or when a decision is waiting on the user, and recurring check-ins should include an exact time zone and end date.
Human approval remains the control point
Dots can be instructed to draft messages, but not send them, or to ask before changing shared files. Built-in approval checks and app permissions still apply, and extra custom rules can tighten boundaries further. A common example is requiring approval before any customer-facing message is sent.
Good briefs separate facts from questions
If a design review moves from Friday to Wednesday, the update should flag the revised date, identify the owner, and show which downstream work is now at risk. It should distinguish confirmed information from unresolved issues and link back to the provided materials. The result is a draft for review, not an unquestionable source of truth.
Progress and schedules need active oversight
In the desktop app, users can inspect activity to review progress, files, and requests, then add missing context or redirect the task without starting over. A separate scheduled view shows recurring jobs, timing, instructions, and destinations for results. That split matters because pausing the main task does not automatically cancel a saved schedule.
Cloud work and local computer access are different
A dot’s cloud computer is separate from a user’s own laptop or desktop. Cloud-based work may continue while the user’s device is off, but any task requiring the local machine depends on that computer being online and separately authorized. Access to messaging channels, app connections, and personal computer control are distinct permission choices.
Availability is still limited
Dots are rolling out gradually, with access varying by plan, region, and workplace administration settings. Some users may need an administrator to enable the feature before it appears. Where dots are not yet available, the same operational model still applies to current AI workflows: one narrow job, clear sources, useful updates, and human review.
CONCLUSION
Dots are most effective as tightly scoped assistants that maintain momentum on a specific task while escalating decisions to a person. Their value depends less on autonomy than on precise instructions, explicit monitoring rules, and firm approval boundaries.
Ask a question