---
title: "ChatGPT Dots vs Grok Bot: work surface or production agent"
description: "An operational comparison of ChatGPT Dots and Kiia's Grok Bot across channels, permissions, human handoffs, and systems of record."
author: "Carlos García"
published: 2026-09-30
updated: 2026-09-30
language: en
human_url: "https://kiia.cloud/blog/dots-vs-grok-bot/"
---

> **In 60 seconds:** choose Dots when you want an always-available personal agent inside the ChatGPT ecosystem, with continuity across ChatGPT, Slack, and Teams, a cloud computer, plugins, and OpenAI's controls. Evaluate Grok Bot when the problem starts with a repeated operational responsibility: a specialist receives work, maintains a backlog, hands evidence to a person, and returns the approved result to the system of record. In this article, **Grok Bot means Kiia's product/assistant, not xAI's Grok model**. Kiia has not published a complete matrix of Grok Bot channels, connectors, permissions, or pricing; confirm those points with the product team. This decision does not come from a benchmark. It comes from where state lives, who can authorize an action, and how the team recovers an exception.

## This is not a model shootout

“Which one is smarter?” distracts from the actual decision. An agent may write excellent copy and still fail in production because it missed an exception, changed the wrong record, or left nobody responsible for reviewing the work.

[OpenAI introduced Dots on September 29, 2026](https://openai.com/index/introducing-dots/) as always-on agents powered by GPT-6 Astra. Each Dot has a cloud computer and browser, works with connected apps, and keeps context as the conversation moves between channels. It is a general surface that follows a person and their projects.

Kiia's published Grok Bot pattern starts at the other end. It first defines a specialist, its outcome, sources, queue, and approval point. Only then does it decide which tools are needed. Our field guide to [Grok Bot in production](/blog/grok-bot-specialist-agents-in-production/) and the [Financio example](/blog/financio-grok-bot-expense-ledger-template/) show this operational approach. They do not provide a complete public list of channels or integrations, so we do not infer one.

## Operational comparison

| Decision | ChatGPT Dots | Kiia's Grok Bot |
| --- | --- | --- |
| Starting point | A personal agent inside ChatGPT that keeps work moving between conversations | A repeated job assigned to a specialist with defined inputs, outputs, and boundaries |
| Published channels | ChatGPT on desktop, web, and mobile; Slack and Teams; calls inside ChatGPT | Desktop assistant. Other channels are not published: confirm with product |
| Continuity | Memory, notes, and relevant context across conversations and channels | The published pattern uses a brief, backlog, and system of record so state does not depend on chat |
| Tools | Cloud computer and browser, plugins, and an authorized connected computer | Specific connectors and integrations are not published: confirm with product |
| Permissions | ChatGPT app permissions, built-in rules, Custom Rules, and auto-review | Kiia describes least privilege and approval as design principles; exact implementation is not published |
| Human handoff | Dots asks for decisions, shows activity, and hands off actions based on automatic review | An owner reviews evidence and approves the exact change before it reaches the official system |
| System of record | The primary Dot works through connected apps; OpenAI places deep SoR integrations in the specialist dots preview | The system of record is explicit in the operating pattern; direct writes and available connectors must be confirmed |
| Price, ROI, and benchmarks | Not compared in this article | Not published for this comparison; confirm with product |

The table does not claim feature parity. It also separates an available product from an announced direction. OpenAI says the primary Dot can use connected apps. Its **specialist dots**, with their own identity, credentials, and deep system-of-record integrations, are in preview and focused enterprise pilots. Do not design a critical rollout today as though every preview capability were generally available to every account.

## When to choose Dots

Dots fits when the main friction is maintaining continuity around one person. An owner can begin an analysis in ChatGPT, share context with the team in Slack, and receive a pending decision in Teams without creating a new agent or resetting its memory.

OpenAI's official guide to [how Dots work](https://learn.chatgpt.com/docs/dots) confirms four useful parts of that experience:

1. A Dot continues working on its cloud computer while the person's device is off.
2. The same Dot is available through ChatGPT, Slack, Teams, or a call.
3. Plugins use connected accounts and their existing permissions.
4. Actions that affect accounts or share information go through review, which determines whether the Dot can proceed, needs approval, or must hand the step to the person.

That makes Dots a natural candidate for research, document preparation, project follow-through, and personal coordination. Its value is a continuous surface close to the user.

Do not mistake convenience for a complete operating policy. Custom Rules can allow, block, or require approval for particular actions, but the team still has to decide which record is official, who handles an exception, and how the result is audited.

## When to evaluate Grok Bot

Grok Bot fits when the unit of design is a responsibility, not a conversation. Examples include turning approved briefs into issues, preparing receipts for review, or maintaining a content-exception queue.

The specialist gets a narrower definition:

- which event or queue starts the work;
- which sources it may read;
- which artifact it must deliver;
- which actions are prohibited;
- which person resolves ambiguity;
- which system holds the current state.

Chat remains useful for steering and correction. It should not be the official backlog. When a case stops, another person should be able to open the work tracker, see the source, understand the state, and continue without reconstructing an entire conversation.

This approach is useful when a process crosses a real boundary: operations to finance, product to engineering, or content to publishing. However, Grok Bot's commercial and technical details are not documented enough outside Kiia's blog. Before committing to a design, ask the product team to confirm channels, connectors, isolation, retention, permission model, limits, and support in writing.

## Control: a rule does not replace a system

Dots publishes a concrete control layer. Custom Rules complement its built-in rules, and auto-review checks sensitive actions against instructions, rules, and safety requirements. OpenAI also says that some tasks must always remain with the person and that Dots can make mistakes.

For Grok Bot, Kiia has published operating principles, not a complete product-control specification. The design should therefore assume only what can be verified: least privilege, reviewable output, and explicit approval. If a restriction matters, enforce it in the account, API, or destination system too. Writing “do not publish” in a prompt is not enough.

For either product, assign autonomy by action:

| Action | Recommended starting mode |
| --- | --- |
| Read approved sources and summarize | Automatic, with source records |
| Propose a change or prepare a draft | Automatic, visible in a queue |
| Write a reversible record | Approve the exact payload and keep a log |
| Send, publish, pay, or change permissions | Named human owner and system control |
| Delete data, merge code, or deploy to production | Outside the pilot until review, recovery, and explicit authorization exist |

## Handoffs: design the exception before the happy path

A handoff is not a message that says, “I need help.” It should include the case, source, attempted work, pending decision, and prepared action. It also needs a responsible person or role and an expected response time.

With Dots, that handoff can arrive in the agreed channel and remain visible in Activity. In the Grok Bot pattern, it is better represented as a backlog state such as `needs review`, `blocked on data`, or `ready to approve`. The labels may differ. What matters is that the case can be inspected outside the agent's memory.

Test three exceptions before expanding a pilot: a required field is missing, two sources disagree, and the action exceeds available permission. If the team does not know who responds or how work resumes, it does not yet have a production workflow.

## The system of record matters more than the channel

Slack, Teams, and ChatGPT are good places to talk. A CRM, ERP, CMS, or project tracker is usually a better place to hold the state the business governs.

Ask where the team will look tomorrow for the current version. If the answer is “find the latest message,” the workflow is incomplete. A sound pattern keeps these elements in the official system:

- a stable case identifier;
- the minimum source or evidence;
- state and responsible owner;
- the proposed and approved changes;
- the date and actor for each transition.

OpenAI previews specialist dots with deep connections to systems of record, but describes that offering as a preview. Grok Bot is framed around this pattern, though its public connectors are not enumerated. In both cases, validate real writes, idempotency, and recovery before promising end-to-end automation.

## Two common decisions

**A founder wants continuous help with research, documents, and coordination.** They already work in ChatGPT and Slack, switch topics throughout the day, and want the same context to follow them. Dots is the first candidate. Start with scoped apps, drafts, and a Custom Rule that reserves external commitments for approval.

**Operations needs to process a weekly queue with exceptions and traceability.** Every case needs a source, state, owner, approval, and a write to the CRM or ERP. Grok Bot may be a candidate if the product team confirms the required connectors and controls. OpenAI's enterprise specialist dots preview is also worth evaluating, without treating an announced capability as general availability.

## Decision checklist

Answer these questions in writing before choosing:

1. Is the unit of work one person's changing agenda or a repeated business queue?
2. Which channel does the user need, and is it publicly supported by the product?
3. What is the system of record, and which exact operation must the agent perform?
4. What minimum permissions does it need to read and write?
5. Which actions must always wait for a person?
6. How is an exception displayed, assigned, and resumed?
7. Which claims depend on a demo or preview, and which are available in the actual account?

If the answers favor personal continuity and integrated channels, test Dots. If they favor a specialist role with a backlog, handoffs, and governed state, evaluate Grok Bot with the product team. If the workflow requires strict contracts, customer-level isolation, deterministic logic, or regulated auditability, a custom agent is probably the better choice.

## Sources and currency

We verified Dots capabilities on **September 30, 2026** using official OpenAI sources:

- [Introducing dots](https://openai.com/index/introducing-dots/), published September 29, 2026.
- [Meet dots](https://learn.chatgpt.com/docs/dots), ChatGPT documentation reviewed September 30, 2026.

For Grok Bot, we used only product context already published by Kiia in [Grok Bot in production](/blog/grok-bot-specialist-agents-in-production/) and [Financio](/blog/financio-grok-bot-expense-ledger-template/). Additional channels, connectors, pricing, isolation, and detailed technical controls remain **not published / confirm with product**. We did not use ROI figures or benchmarks to make the decision.

## Frequently asked questions

### Do ChatGPT Dots and Grok Bot use the same kind of agent?

They should not be compared as equivalent models. Dots is an OpenAI product with a personal surface across ChatGPT, Slack, and Teams. In this article, Grok Bot is Kiia's assistant for organizing specialists around repeated jobs, permissions, a backlog, human review, and systems of record.

### Which one fits a team that works in Slack or Teams?

Dots publicly supports talking to the same agent in ChatGPT, Slack, and Teams. Kiia has not published an equivalent channel matrix for Grok Bot, so that coverage must be confirmed with the product team. A channel also does not replace permission boundaries or an official destination for the work.

### Can either product act without approval?

Not as a general rule. Start with reading, analysis, and drafts. Keep sends, payments, deletion, permission changes, publishing, and production work behind explicit approval and system controls. Assign autonomy by action and risk, not to the entire agent.