OpenAI dots: what Technical Writers need to know

OpenAI has announced dots, a new type of always-on AI agent that can continue working between conversations.

Most AI assistants wait for a prompt. A dot can be given an ongoing area of work to monitor.

A dot has its own cloud computer and browser. It can use connected applications, carry out scheduled work, respond to certain events, and coordinate work with other AI agents. It also retains notes about preferences, decisions, and ongoing responsibilities.

OpenAI is initially rolling out dots to eligible Pro, Business Premium, and Enterprise users. See OpenAI’s press release: Introducing dots.

We’ll be including dots in our new Managing documentation projects with AI course. Below is an initial summary of its relevance to, and potential for, Technical Authors.

From individual tasks to ongoing documentation work

You might ask an AI assistant to draft release notes from a set of completed tickets.

A dot could have a broader responsibility:

Monitor approved product changes, identify those that affect customers, maintain the documentation-readiness register, and prepare proposed updates for review.

Other potential responsibilities for dots include:

• Checking whether documentation is ready for an upcoming release.
• Monitoring support tickets for recurring documentation problems.
• Comparing product changes with the existing help centre.
• Preparing a weekly report on overdue reviews and unresolved questions.
• Flagging contradictions between policies, procedures, and customer-facing content.
• Coordinating reviews without publishing unapproved material.

Defining the responsibility

The dot needs to know which sources are authoritative, what counts as a documentation change, who can approve it, and which actions it must never take without permission.

This means a specification should cover:

• The documents, systems, and communication channels it can use.
• The events or schedule that trigger work.
• The decisions it can make.
• The actions that require human approval.
• The evidence it should retain.
• The people it should contact when information conflicts.
• The conditions under which it should stop.

OpenAI provides users with controls for deciding whether a dot can act without asking, act following a direct request, ask for approval, or hand the action back to the user.

These controls sit alongside application permissions and built-in safeguards. However, OpenAI’s guidance says these rules can still be applied incorrectly,

Any important work needs to be reviewed.

Gathering evidence of success

An agent can finish a task without achieving the intended result.

For example, it might prepare a documentation update but use an obsolete product specification. It might send a review request to the wrong person.

Documentation teams will need measures to check things such as. Did it identify the right changes? Were its sources current? Did the reviewer accept the draft?

See: A completed run does not confirm that the requested result was achieved or delivered.

Memory needs managing

A dot uses conversation context, relevant ChatGPT memory, and its own saved notes. Those notes can record preferences, decisions, and ongoing work.

But they are not a complete transcript.

There is likely to be a need to carry out maintenance. Someone will need to correct obsolete assumptions, review changing responsibilities, and check whether information learned in one context is appropriate to use in another.

Stopping work needs care

OpenAI states that pausing the main dot does not automatically stop delegated tasks or cancel future scheduled work.

Any “offboarding” should cover tasks, schedules, connected applications, stored notes, accounts, and access permissions.

The opportunity for Technical Writers

Dots point towards a form of documentation management where an agent watches the work continuously rather than waiting for someone to remember the documentation.

That could reduce missed updates and administrative effort.

It could also automate confusion if the responsibility, source material, permissions, and review process are poorly defined.

Technical Writers will need to identify reliable sources, resolve ambiguity, define approval processes, and decide what readers need to know. The need to make responsibilities explicit will be important.

Note: We used ChatGPT to help gather and summarise the information about dots.