TL;DR: OpenAI Dots are always-on agents in ChatGPT that work in the background, on their own cloud computer, across thousands of connected apps. For one person’s admin, research and follow-up work they are a strong, low-cost place to start, and a custom AI agent still wins when the work is a team process rather than a personal one, when it depends on your own systems and business rules, when the result has to be the same every time, or when the data has to stay inside your own environment. Most businesses will end up using both.
Contents
- What OpenAI Dots Are
- What a Custom AI Agent Is
- OpenAI Dots vs Custom AI Agents at a Glance
- Where OpenAI Dots Are the Right Choice
- Where a Custom AI Agent Still Wins
- Data, Control and Canadian Businesses
- How to Decide: A Two-Week Test
What OpenAI Dots Are
OpenAI Dots are always-on AI agents built into ChatGPT. OpenAI announced them at DevDay on September 29, 2026, and describes them as agents that get to know what matters to you and keep working on your behalf.
Three things separate a dot from a chatbot:
- It keeps working when you leave. Each dot has its own cloud computer and can run several projects at once, 24 hours a day, without you starting each step.
- It connects to your tools. Through OpenAI’s plugin ecosystem a dot can reach more than 4,000 apps, and you can talk to it in ChatGPT, Slack or Teams.
- It learns from feedback. Over time it adapts to how you like work done.
Examples given at launch include investigating a bug someone mentions in Slack, testing a fix and opening a pull request, rerunning an analysis, revising a proposal, turning transcripts into clips and drafting social posts.
At launch, Dots are rolling out to ChatGPT Pro and Business Premium users in eligible markets. Enterprise, Edu and Healthcare workspaces can try a beta once an admin enables it. Pro access excluded the European Economic Area, Switzerland and the UK at launch.
The first dot is included in the plan, and dot conversations do not count against plan limits for the first month.
What a Custom AI Agent Is
A custom AI agent is software built around one specific job in your business. It uses a language model for the reasoning, but the steps, the systems it can touch, the rules it follows and the checks on its output are designed for that job and nothing else.
Examples: an intake agent that reads every new enquiry, checks it against your service area and pricing rules, and books qualified ones into the right calendar. A quoting agent that pulls costs from your own price book. A reconciliation agent that matches invoices to purchase orders and flags only the exceptions.
The difference is not intelligence. A custom agent may use the same class of model a dot does. The difference is that a dot is a capable generalist you direct, and a custom agent is a specialist built to do one process the same way every time, whoever is on shift.
OpenAI Dots vs Custom AI Agents at a Glance
| OpenAI Dots | Custom AI agent | |
|---|---|---|
| Built for | One person’s work, across many tasks | One business process, done consistently |
| Setup | Minutes, inside ChatGPT | Days to weeks of design and build |
| Cost to start | First dot included in an eligible plan | Build cost, then running costs |
| Knows your business | Learns from your feedback and connected apps | Your rules, data and systems are designed in |
| Consistency | Varies with how it is asked | Same steps and checks on every run |
| Who it works for | The person who owns it | A team, a queue or a customer-facing channel |
| Systems it reaches | Apps with a plugin, under your own permissions | Any system you can connect, including internal ones |
| Where it runs | OpenAI’s cloud | Where you choose, including in Canada |
| Best first use | Admin, research, follow-ups, drafts | Intake, quoting, reconciliation, routing |
Where OpenAI Dots Are the Right Choice
Start with a dot when the work belongs to one person and a good result is easy to recognize.
- Personal admin. Sorting an inbox, chasing replies, preparing meeting notes, keeping a task list current.
- Research that runs in the background. Watching competitors, collecting supplier quotes, summarizing what changed in a market this week.
- First drafts. Proposals, posts, internal updates, where a person reviews before anything goes out.
- Small builds. Simple integrations or reports that one person owns and maintains.
The economics favour it. If you already pay for an eligible plan, the first dot costs nothing extra, and there is no project to scope. For many owners and managers that removes a few hours of low-value work a week, which is a real return for no build cost.
Where a Custom AI Agent Still Wins
A dot works for the person who owns it, with that person’s access. That is exactly right for personal work and the wrong shape for a business process. A custom agent is the better fit in five situations.
1. The work is a team process, not a personal task
Customer intake, dispatch, quoting and support queues run on behalf of the business, not one employee. They need to keep running when someone is on holiday, hand work between people, and leave a record anyone can audit. An agent tied to one person’s account is a single point of failure for work the whole team depends on.
2. It depends on your own systems and rules
A dot reaches the apps that have a plugin. Many businesses run on a job-management system, an industry platform, a shared spreadsheet or a database with no plugin at all.
The rules that make an answer correct, such as your service area, discount limits or approval thresholds, live in your head and your documents, not in a general tool. A custom agent has those rules built in and is tested against them.
3. The result has to be the same every time
A generalist agent decides how to approach each task, which is its strength for open-ended work and a weakness for a process. When a quote, a compliance check or a customer reply has to follow the same steps on every run, you want those steps fixed in the design and checked automatically, not re-derived from a prompt.
4. It faces customers or moves money
OpenAI itself says dots can make mistakes and that consequential work should be reviewed. That is sound advice for any agent.
With a custom agent you decide where the human review sits, which actions are impossible rather than merely discouraged, and what happens when the agent is unsure. For anything that sends to customers or touches payments, that design work is the product.
What that looks like in practice
Take a home-services company with three office staff sharing one enquiry inbox. Each could hand the inbox to their own dot, but each dot would sort leads a little differently, and nobody’s dot covers the inbox while its owner is away.
Kaxo’s AI agent development replaces that with one intake agent owned by the business. It reads every enquiry, applies the company’s own service area and job-size rules, books qualified jobs into the shared calendar, and hands anything unusual to a person with the reason attached. The dots stay, doing each person’s own follow-ups and research.
5. You need to know what it costs per job
A custom agent’s running cost can be measured per task and controlled by choosing the right model for each step. OpenAI has said more dots and higher workloads will be priced later but has not published those prices. If an agent is going to handle hundreds of jobs a week, cost per job is a number you want before you depend on it.
Data, Control and Canadian Businesses
OpenAI has put real controls around Dots. Content in Business, Enterprise and Edu workspaces is not used for training by default. A dot’s cloud computer is separate from yours unless you connect it.
Custom rules let you allow an action, require approval for it, or block it, and some sensitive actions, such as changing passwords, always stay with you.
What a dot does not change is where the work happens. It runs on OpenAI’s infrastructure, with access to whatever you connect it to. For most general office work that is acceptable.
For a Canadian business handling health information, client financial records or data with contractual residency requirements, the question is not whether the controls are good but whether that data may leave your environment at all.
A custom agent can be built to run where your data has to stay, with access limited to exactly the systems the job needs. If you are unsure which side of that line your data falls on, settle it before connecting any agent to it. Our guide to AI security and compliance for Canadian businesses covers the questions to ask, and Canadian data sovereignty explains when residency is worth paying for.
How to Decide: A Two-Week Test
You do not have to choose up front. The cheapest way to find out where a general agent is enough is to try one.
- Pick one bounded job. Choose something recurring, low-risk and easy to check, such as weekly supplier follow-ups or a research digest.
- Give a dot that job for two weeks. Set custom rules so it asks before sending anything outside the business.
- Keep a simple log. Note the time it saved, every correction you made, and every point where it could not reach a system or did not know a rule.
- Read the log. If the corrections are rare, keep the dot and give it the next job. If they cluster around the same systems, rules or hand-offs, you have just written the brief for a custom agent.
That last point is the useful one. The places where a capable generalist keeps stumbling are the places where your business is specific, and specific is where a custom build earns its cost. For a fuller look at the trade-offs, see our build vs buy framework for AI agents, and if the test points to a custom build, how to choose an AI agent development company.
If you want a second opinion on which of your processes belong with a dot and which need their own agent, talk to us about AI agent development. We run agents on our own operations and measure what they do, so the advice comes from production rather than a demo.
Sources: OpenAI, “Introducing dots” and DevDay 2026 announcements (September 29, 2026); launch coverage by 9to5Google, MediaNama and Techloy. Product details are as announced at launch and may change as the rollout continues.
Frequently Asked Questions
What are OpenAI Dots?
Dots are always-on agents inside ChatGPT that OpenAI launched on September 29, 2026. Each dot runs on its own cloud computer, keeps working on assigned tasks after you log off, connects to other apps through plugins, and can be reached in ChatGPT, Slack or Teams. OpenAI says they run on GPT-6 Astra and learn from feedback over time.
Who can use OpenAI Dots?
At launch, Dots are available to ChatGPT Pro and Business Premium users in eligible markets, and Enterprise, Edu and Healthcare workspaces can try a beta once an admin turns it on. Pro access excluded the European Economic Area, Switzerland and the UK at launch. The first dot is included in the plan.
How much do OpenAI Dots cost?
Your first dot is included in a Pro or Business Premium plan at no extra cost. OpenAI has said more dots, and options to change their speed and monthly workload, will come later, but it has not published prices for them. Usage terms for each plan are due after the first month.
Are OpenAI Dots a replacement for custom AI agents?
For general office work run by one person, often yes. A dot is a capable generalist that works inside the permissions of the person who owns it. A custom agent is still the better fit when the work runs across a team, depends on your own systems and rules, has to produce a consistent result every time, or touches data you must keep under your own control.
Is my data safe with OpenAI Dots?
OpenAI says content in Business, Enterprise and Edu workspaces is not used to train its models by default, that a dot's cloud computer is separate from yours unless you connect it, and that you can set rules for which actions a dot may take alone, which need approval and which are blocked. It also says dots can make mistakes and consequential work should be reviewed. Whether that meets your obligations depends on the data involved and where it has to live.
Should a small business try OpenAI Dots first?
If you already pay for an eligible plan, yes. Give a dot one bounded, low-risk job for two weeks and measure it. What it struggles with is the most useful brief you can write for a custom agent, because it shows exactly where a general tool runs out.