What Is a Workflow in OpenAI Dots? Goal, Tools, Trigger, Rules
Within hours of OpenAI launching dots on September 29, 2026, people were talking about “dot workflows”. There is no feature with that name. There is no canvas, no boxes and arrows, and nothing to drag.
What exists is four separate parts that together behave like a workflow: a goal, connected tools, a trigger and rules. This post explains each, shows where OpenAI’s terms map onto them, and takes one marketing workflow apart. It is written as of launch day. For what dots are in the first place, see OpenAI dots for marketing.
The four parts
| Part | The question it answers | OpenAI’s term |
|---|---|---|
| Goal | What is the dot responsible for? | The goal you give your dot, and skills |
| Tools | What can it reach? | Plugins and their MCP servers |
| Trigger | When does it run? | Automations, on a schedule or an event |
| Rules | What may it do alone? | Custom Rules |
OpenAI’s launch description covers three of these in one sentence: give your dot a goal, connect the apps it needs, and define what it can do on its own. The fourth, the trigger, comes from automations.
Goal: what the dot is responsible for
A goal is a standing responsibility, not a single request. “Write a launch email” is a request. “Keep our launch material current whenever the product changes” is a goal.
When the same job has a fixed procedure, it becomes a skill. OpenAI’s plugin documentation defines a skill as a folder containing a SKILL.md file and, when needed, supporting scripts, references, templates or assets. In plain terms it is the written version of how your team does something: check these three things, in this order, and report in this format.
Tools: what the dot can reach
A dot works through the apps you connect. Each connection is a plugin, and the part of a plugin that does the work is usually an MCP server: a list of tools with defined inputs and outputs, and a sign-in.
The same documentation lists four shapes a plugin can take.
| Shape | What it gives the dot |
|---|---|
| Skills only | A procedure, using tools it already has |
| MCP server only | Tools, with no extra instructions |
| Skills and MCP server | A guided procedure using custom tools |
| MCP server and interface | Tools with a visual panel |
The tool list is the most important decision in the whole workflow, because it sets the ceiling. A dot cannot do what no tool allows. When you evaluate a marketing tool’s connector, read the tool list before anything else. The checklist for that is in does it have an MCP server?.
Trigger: when it runs
Automations decide the timing. There are two kinds.
Schedule. Every Monday at nine. The first weekday of the month. This works with any connected tool.
Event. Something changed in a connected app. OpenAI announced support for the proposed MCP Events specification at DevDay, so a supported plugin can start an automation when, for example, a new task lands on a project board. The keynote recap covers the announcement.
Two cautions on events. The specification is still a proposal, and it only works with plugins that implement it. As of September 2026, Marqeable’s connector does not send events. A dot reaches Marqeable on a schedule, or when an event from another app starts the run.
Rules: what it may do alone
There are two layers. Each connected plugin has a permission level: always ask, allow read-only tools, allow low-risk tools, or allow all. On top of that, Custom Rules are sentences you write about an action, each set to one of four behaviors: take action without asking, take action when you say so, ask before taking action, or hand off to you. OpenAI’s auto-review checks consequential actions before they proceed, as described by VentureBeat.
A workable starting point for marketing:
| Action type | Setting |
|---|---|
| Reading anything | Plugin permission: allow read-only tools. No rule needed |
| Creating a draft | Ask before taking action, then take action without asking once you trust the output |
| Changing account settings | Ask before taking action |
| Sending, publishing, deleting | Hand off to you, or better, do not connect a tool that can |
One workflow, taken apart
Here is a single marketing workflow written as its four parts. It uses Marqeable’s MCP server for the tools.
| Part | Setting |
|---|---|
| Goal | Every Monday, tell me whether next week’s marketing matches our current focus, and draft what is missing |
| Tools | get_business_context, get_content_calendar, list_campaigns, create_campaign_draft |
| Trigger | Schedule: Mondays, start of the working day |
| Rules | Plugin on allow read-only tools. Draft creation: take action without asking. Settings changes: ask before taking action |
What happens on a Monday: the dot reads the business context to learn the current offer and focus. It reads the calendar for the next four weeks. It lists campaigns to see what is already planned. If there is a gap, such as a product date with nothing scheduled around it, it creates a campaign draft and sends you a message with the link. Nothing is sent to a customer. You open the draft in Marqeable and decide.
We ran this on our own dot on September 30, 2026. In chat, it made the four reads, reported one calendar entry with no campaign linked, and on request created one campaign draft and returned the link. A scheduled read-only version then ran unattended and reported back into the dot’s chat, about five minutes after its set time. The run is written up in connecting Marqeable to a dot.
What a dot workflow is not
In marketing, “workflow” already has a meaning: a program that contacts enroll in, where each person moves through steps on their own timing. HubSpot calls these workflows. Marqeable calls them automations.
A dot workflow is not that. It is work done by one agent. It has no per-contact state, no enrollment and no delivery. The two fit together, with the dot preparing and the automation delivering, and the reasoning is in dots do not replace your marketing automation. The same distinction for an older generation of tools is in HubSpot workflows vs AI agents.
Frequently asked questions
Do OpenAI dots have a workflow builder?
No. As of September 2026 there is no canvas or flowchart. You describe the work in plain language. What people call a workflow is made of four parts: the goal you give the dot, the apps and tools you connect, the automation that decides when it runs, and the rules that decide what it may do without asking.
What is the difference between a skill and an automation?
A skill is the procedure: a folder with a SKILL.md file of instructions and any supporting material. An automation is the timing: it runs work on a schedule or when an event arrives. A skill says how the Monday check is done. An automation says it happens every Monday.
What is MCP Events?
MCP Events is a proposed addition to the Model Context Protocol that lets a connected app tell the assistant that something changed. OpenAI announced support for it at DevDay 2026 so that a supported plugin can start an automation, for example when a new task appears on a project board. It only works with plugins that implement it.
How is a dot workflow different from a marketing automation?
A dot workflow is a piece of work the agent does, such as checking a calendar and drafting a campaign. A marketing automation is a program that contacts enroll in, one by one, and that sends messages to each of them on its own timing. The first produces drafts and decisions. The second delivers messages.
The bottom line
When someone says dot workflow, ask which part they mean. The goal is what you want. The tools are what the dot can reach. The trigger is when it runs. The rules are what it may do alone. Get those four written down for one recurring job and you have a workflow, with no builder required.
See what the drafts become: Marqeable’s agents build and run your campaigns, run behavior-triggered automations across text and email, answer and qualify every visitor with AI website chat, and tie revenue to the exact message with attribution.
Marqeable runs your campaigns, answers every visitor, text, and email in seconds, and turns them into booked jobs and meetings - even at 9pm on a Saturday. We’re in private beta with a small early cohort. Get early access
