What you are really choosing
dots is a general agent product from OpenAI, launched in September 2026. A dot is an always-on agent that works from its own cloud computer, so it keeps going when your laptop is closed. You give it a goal in ChatGPT, Slack or Microsoft Teams, connect the apps it needs from a catalogue of thousands, and set Custom Rules for what it may do alone, what needs your approval and what it must never do. Between jobs it does read-only research in the apps you have connected, and it learns your preferences from feedback over time.
dots comes with ChatGPT’s Pro, Business Premium and Enterprise plans in the markets where it has launched, and the first dot is included. For companies, OpenAI is piloting specialist dots, each with its own identity and access to the systems it needs.
If you are looking at dots for business automation, you are probably asking a wider question. Which kind of tool should take on the recurring work in your business, and what should it be able to do?
The alternatives fall into a few groups. Chat assistants help a person work faster. Rule-based automation tools connect apps with trigger and action rules. Assistants inside office suites, like Microsoft Copilot, help inside the apps your team already uses. Business agent products take on recurring jobs across your systems with approvals. Custom built agents are made by your own team.
Each fits a different kind of job. The rest of this guide helps you match the job to the tool and gives you the questions to ask of any product, dots included.
Start with the job, not the product
Comparisons of agent products tend to turn into feature lists. Features change often and are hard to check from the outside. The jobs your business needs done change much more slowly.
Before you compare any products, write down the work you want covered. For each job, note four things.
- How often it happens. Once, weekly, daily or whenever something changes.
- Where it lives. The inbox, the calendar, the CRM, the store, the books or a shared spreadsheet.
- Who decides. Which steps a person must approve, and which are routine.
- What a mistake costs. A slightly awkward email, a lost lead, a wrong payment.
A job that is frequent, lives in your systems, has clear rules and costs something when it slips is a good candidate for an agent. A job that is rare, varied or very costly to get wrong may be better left with a person and a good assistant.
The kinds of alternatives
Chat assistants
A chat assistant such as ChatGPT is a tool a person opens and works with. It drafts, summarizes and explains. It is a strong choice when a person owns the task and wants help with each step. It is a weaker fit for work that has to happen on time when nobody is at the keyboard.
Rule-based automation
Tools like Zapier connect apps with trigger and action rules. When a form is filled in, add a row to a spreadsheet and send a message. These are reliable for fixed steps. They struggle when a case does not fit the rule, because they follow the steps whatever the situation.
Assistants inside office suites
Microsoft Copilot is an assistant inside Microsoft’s apps. Tools of this kind help people write, summarize and find things within the documents, email and meetings they already use. They suit organizations that live in one suite and want help where the work already happens.
Business agent products
These are agents built around recurring business jobs. You delegate a job, such as lead follow-up or collections, and the agent works on it over weeks in your tools, with approvals on the actions that matter.
Custom built agents
Some teams build their own agents on a model provider. This gives the most control and asks the most of you. Someone has to build it, monitor it, fix it and own its mistakes.
Options side by side
The table describes each kind of option by what it asks of you and what to check. It does not rate any product, and the cells are questions rather than specs.
| Option | Kind of tool | Good fit when | What it asks of you | What to check |
|---|---|---|---|---|
| dots | General agent product from OpenAI | Check its current documentation against your job | Instructions and access to whatever it needs | Does it support the access, approvals and record your job needs? |
| Chat assistant | A person works with it in a chat | A person owns the task and wants help | Time from the person doing the work | Does the work still slip when that person is busy? |
| Rule-based automation | Trigger and action rules between apps | The steps are fixed and rarely change | Someone to build and maintain rules | What happens to the cases that do not fit? |
| Office suite assistant | Help inside the apps your team uses | Work happens inside one suite | Licenses and data access policy | Can it act across tools outside the suite? |
| Business agent products | Agents built for recurring business jobs, with approvals before anything that matters | Recurring jobs in a small business | Delegating a job and approving key actions | Which jobs it covers, and which approvals you want on |
| Custom built agents | Agents your team builds and runs | You need full control and have engineers | Build time, upkeep and monitoring | Who owns it when it breaks? |
Most businesses end up using more than one row. A team might use a chat assistant for writing, a rule-based tool for simple handoffs and an agent for the follow-up that keeps slipping.
Holding each option up against dots
dots sets a clear bar on a few points. Ask every option how it compares.
| Check | Question to ask |
|---|---|
| Your current plan | Do you already have ChatGPT Pro, Business Premium or Enterprise, and has dots launched where you are? |
| Rules | dots lets you set what it does alone, what needs approval and what it must never do. Can the option draw the same three lines? |
| Running alone | A dot keeps working from its own cloud computer with your laptop closed. Does the option keep going when nobody has it open? |
| Between jobs | A dot does read-only research in your connected apps between jobs. Do you want that, and can you limit it? |
| Learning | dots learns your preferences from feedback. Can you see and correct what each option has learned? |
| Identity | Specialist dots with their own identity are still a pilot. Whose name does each option act under in your systems today? |
Access, the record, data and exit apply to any product, and the full checklist has them. Ask for answers in writing, then check them in the pilot.
Running a fair comparison
Demos are built to impress. A pilot shows what happens in an ordinary week. To compare options fairly, keep everything but the tool the same.
- Pick one real job. Lead follow-up and invoice reminders are good choices because the results are easy to see.
- Write one brief. Use the same goal, rules and approvals for every option you test.
- Connect the same tools. Give each option the same scoped access to the same inbox, calendar or CRM.
- Run it for a few weeks. Long enough to see quiet days, busy days and at least one odd case.
- Read the record. Check what each one did, what it asked you about and what it missed.
- Score the same things. On time, accurate, inside the rules, easy to correct.
The option that wins is rarely the one with the best first day. It is the one you trust most by the end of the third week.
Choosing
If your work is varied and a person drives each task, a chat assistant or a general agent may be all you need. If the work is fixed steps between two apps, a rule-based tool may do it cheaply. If the work is recurring, lives across your systems and keeps slipping when people are busy, look closely at business agent products and at whether building your own is worth the upkeep.
Whatever you choose, keep a person in charge of the decisions that shape how your business is seen and how its money moves. The right tool takes the steady work off your team’s plate and makes it easy to see what was done.
