ChatGPT agent mode and OpenAI dots both let ChatGPT use a computer for you, but they are built for different jobs. Agent mode, launched in July 2025, runs a single task when you ask and then stops. A dot, announced at DevDay on 29 September 2026, is a persistent agent that keeps working in the background, remembers you and can come to you first. This comparison is based on OpenAI's documentation and launch coverage; we have not tested either product hands-on.

What is ChatGPT agent mode?

OpenAI introduced ChatGPT agent on 17 July 2025, combining its earlier browser-using agent, Operator, with deep research. You switch it on from the tools menu in a normal ChatGPT conversation, describe a task, and ChatGPT works through it on a virtual computer with a browser, then reports back.

Key traits at launch:

  • Task-scoped. It works on the job you started in that conversation. When the task is done, so is the agent.
  • Asks before consequential actions. OpenAI said it would request permission before actions such as purchases or sending email, and you could take over the browser.
  • Plan access. It launched for Plus, Pro and Team subscribers with a monthly cap on agent messages. Per OpenAI's help centre, the cap was 40 agent messages a month on Plus and 400 on Pro at last check, and only runs you start (including scheduled ones) count. We could not load the page to confirm current figures, so check your plan's limits in ChatGPT before relying on it.
  • Wide availability. UK sources report that agent mode reached Europe within about six days of its US launch, and it remains one of the few agent tools UK consumers can use today (see our sister site's guide to ChatGPT agent mode in the UK).

What are OpenAI dots?

Dots are what OpenAI calls "always-on agents." Sam Altman described them at DevDay as "remarkably capable, always-on agents built to handle really anything you can think of." Each dot is named and customised by you, and the official docs and launch coverage describe:

  • Its own cloud computer. A Linux environment with Chrome, isolated from other users. You can watch the screen and press "Take over."
  • Background work. A dot keeps going after you close the laptop or the chat, and since 4 November 2026 personal dots keep working on tasks between conversations.
  • Proactivity. OpenAI's example is a dot that noticed an unsent invoice and prepared it for approval. Proactive research is read-only and enforced in code.
  • Memory. A dot receives memories from ChatGPT and forms its own from connected apps. Memory carries across ChatGPT, texting, Slack and voice calls.
  • More channels. ChatGPT on mobile and desktop, voice calls you start, a US texting beta over iMessage and RCS (since 28 October), Slack, Microsoft Teams and Codex.
  • 4,000+ apps through OpenAI's plugin ecosystem.

For the full picture, read our OpenAI dots review.

  1. ChatGPT agent mode launches in the US
  2. Agent mode reaches Europe
  3. OpenAI announces dots at DevDay
  4. dots texting beta and EU Business/Enterprise access
  5. Enterprise, Edu and Healthcare general availability

dots vs agent mode: side-by-side comparison

ChatGPT agent mode OpenAI dots
Launched 17 July 2025 29 September 2026
How you start it Toggle inside a chat Create a named dot once; it persists
Runs when you leave Finishes the current task Yes, keeps working toward goals
Proactive No, waits for you Yes (read-only while proactive)
Memory ChatGPT memory for that chat Own memory plus ChatGPT memories, across channels
Computer Virtual browser per task Persistent cloud computer; optional local computers
Channels ChatGPT ChatGPT, calls, iMessage/RCS beta (US), Slack, Teams, Codex
Approval model Asks before consequential steps Auto-review plus Custom Rules; hard limits on money and passwords
Plans (as of November 2026) Launched on Plus, Pro, Team Pro ($100, $200, $500) and Business Premium only
Usage Monthly agent message cap (40 Plus / 400 Pro, per OpenAI) Separate "deeper work" allowance, extended in month one
UK personal plans Available Not on Pro; Business Premium only

What actually changed between agent mode and dots?

From "do this" to "keep an eye on this"

Agent mode is reactive. You need a task in mind. Dots are designed around the "blank page" problem: a dot can watch named Slack channels, flag a calendar clash or draft a reply to an email you have not opened yet. OpenAI's CFO Sarah Friar called it "the agent that can tap you on the shoulder."

Safety rules became configurable

Agent mode asks for confirmation case by case. Dots add a policy layer. According to eesel's summary of OpenAI's documentation, Custom Rules give four behaviours per action: act without asking, act if pre-approved, ask before acting, or hand off to you. An auto-review step checks planned actions, such as an email's recipient and content, against your rules, and the dot cannot turn it off. Some limits are fixed: password changes and money transfers always come back to you, and saved-card purchases need approval. We compare these rules across vendors in what each agent can do without asking.

Memory became persistent and harder to edit

OpenAI's privacy FAQ says turning off ChatGPT Memory stops new sharing but does not delete what a dot already received. Individual dot memories cannot be viewed, edited or deleted; you reset or delete the dot instead. Agent mode, being task-scoped, carries far less long-lived state. More in agent memory and privacy compared.

The price floor went up

Agent mode came with a $20 Plus plan. Dots start at ChatGPT Pro 100 ($100/month) or Business Premium, which costs $125 per user a month billed monthly or $100 billed annually, with a two-seat minimum. Extra dots and higher capacity are planned as paid add-ons at unannounced prices. Full breakdown: OpenAI dots pricing.

Which one should you use?

Use agent mode if

  • You have occasional, well-defined web tasks: research a purchase, fill a form, compile a spreadsheet.
  • You are on Plus or in the UK, EEA or Switzerland on a personal plan.
  • You prefer an agent that holds no lasting state about you beyond your normal ChatGPT memory.

Use a dot if

  • You want something watching your inbox, calendar and Slack and surfacing what matters.
  • You already pay for Pro or your company has Business Premium.
  • You want to text, call or message the agent in Slack or Teams and have it remember context.

Who should skip dots for now

Reviewers have flagged friction. eesel's review describes a shuttle booking that needed several confirmations and a dot that would not fetch a 2FA code from an inbox it could read. Neoteo found online errands stalled on security checks. If your main need is personal errands rather than work triage, compare Meta Muse vs OpenAI dots before paying $100 a month.

Pros

  • dots keep working in the background and can raise issues first
  • dots remember context across ChatGPT, texting, Slack and calls
  • dots add configurable Custom Rules on top of fixed money and password limits

Cons

  • dots start at $100 a month versus agent mode's $20 Plus entry point
  • Individual dot memories cannot be viewed, edited or deleted
  • No personal Pro access in the UK, EEA or Switzerland

Bottom line

Dots do not replace agent mode so much as sit above it. Agent mode remains the cheaper, more widely available way to hand ChatGPT a single task. Dots add persistence, proactivity, memory and more channels, at five times the entry price and with tighter regional limits. If you are unsure whether you need an always-on agent at all, start with our explainer on AI assistants vs AI agents.