ChatGPT agent mode let ChatGPT do a task for you on its own virtual computer: work through websites, use apps like Gmail, and hand back a report, a deck or a spreadsheet. OpenAI's help page now says agent mode is no longer available, and OpenAI points people to ChatGPT Work for longer tasks and to workspace agents for teams. Where they stop: a person starts or builds each task, signs it in to sites, approves the steps that matter and checks the result. Arkeo agents work another way: each one owns one recurring job in your own software, and we set it up and look after it.
What is ChatGPT agent mode?
OpenAI launched ChatGPT agent in July 2025. Its launch post says ChatGPT can "do work for you using its own computer, handling complex tasks from start to finish." You turned it on from the tools menu in the box where you type. It could click through websites, run code and use apps like Gmail and GitHub, all on a virtual computer of its own.
OpenAI's examples included a brief on upcoming client meetings built from your calendar, and a slide deck on three competitors. A finished task could be set to repeat daily, weekly or monthly. The Plus, Business and Enterprise plans each got 40 agent requests a month. OpenAI said from the start that agent mode was in its early stages and could make mistakes.
What replaced ChatGPT agent mode
OpenAI's help page for ChatGPT agent now says it is no longer available, and the launch post is marked as outdated. Both point to newer parts of ChatGPT.
ChatGPT Work is an agent inside ChatGPT for longer, multi-step work and finished deliverables: research, a document, a spreadsheet, a presentation or a report. On web and mobile it runs in the cloud. When a task needs a website, it uses a web window of its own in the cloud that can sign in to sites, fill in forms and keep working after you close your computer. It pauses when it needs your input, a sign-in or a yes.
Workspace agents are for teams on Business, Enterprise and Edu plans. Someone builds an agent for a repeatable task, connects it to apps like Google Drive, Slack and SharePoint, shares it with the team, and runs it on a schedule or from Slack. OpenAI's examples include a weekly metrics report and a lead outreach agent that updates the CRM.
ChatGPT agent mode vs Arkeo, side by side
| ChatGPT (agent mode, now Work and workspace agents) | Arkeo agents | |
|---|---|---|
| What it is | An agent inside ChatGPT that does a task you hand it, plus agents your team builds | Agents that each own one recurring job in your business |
| Who starts the work | You, task by task, or on a schedule someone sets | The agent, on its own schedule or when something changes |
| Where it works | In the cloud, through connected apps and its own web window | Inside the software your business already uses |
| Who sets it up | You write the task or build the agent, and sign it in to sites | We set it up with you, around how your team works |
| How much watching | It pauses for sign-ins and asks before steps that matter | Your rules decide what it does alone and what goes to a person |
| When your process changes | You update the task or the agent | We update the agent |
How to use ChatGPT Work and workspace agents
Agent mode is gone, so these are the steps for what replaced it, in short. Each one links to OpenAI's own help page.
- 1
Start a task in Work
Open ChatGPT on web or mobile and start a task in Work instead of Chat.
- 2
Describe the outcome
Say what you want done, the website if there is one, and any limits. ChatGPT decides when to use a connected app or its own web window.
- 3
Sign in and approve
When it reaches a sign-in page, sign in through the secure form. Say yes or no when it asks before a booking, a payment or another step that is hard to undo.
- 4
Build an agent for repeat work
On Business or Enterprise, open Agents in the sidebar, describe the workflow, test it, then add a schedule or add it to Slack.
Where ChatGPT agents stop: what is still done by hand
OpenAI's own pages say a person stays close. Its page on web tasks says some tasks "require your review or a final step you complete yourself." It asks before anything hard to undo, like a booking or a payment. Sign-ins to websites last only until they expire, and then someone signs in again.
Some websites block it. The same page says sites can block automated visitors, even when the site works fine for you. When that happens, a person opens the page and does the step by hand.
Workspace agents ask before they write. By default, an agent's changes in connected apps, like sending, editing or posting, are set to "Always ask" during a run, so someone has to be there to say yes.
Agent mode also left a public record of trouble. On OpenAI's community forum, long threads gathered paying users whose agent mode would not start for days or weeks: it sat at "Setting up desktop", or the switch for its web window would not turn on. While a tool like this is down, the work waits until someone notices and does it by hand.
And the work belongs to whoever set it up. A scheduled task or a shared agent keeps doing what it was told. When your process changes, someone has to remember it exists and update it.
Ways to get AI doing work in your business
Most owners end up with one of these, or more than one side by side.
ChatGPT Work, task by task
- How it works
- A person hands ChatGPT a task in Work and checks what comes back.
- Good for
- One-off research, meeting prep, decks and spreadsheets.
- Where it breaks
- It pauses for sign-ins and approvals, some websites block it, and some tasks end with a last step a person does.
Workspace agents
- How it works
- Someone on a Business or Enterprise plan builds a shared agent, connects apps, and sets a schedule or adds it to Slack.
- Good for
- Teams with someone who can build and test agents.
- Where it breaks
- Changes in your apps ask a person first by default, and someone owns the agent: testing it, updating it when the process changes, and checking its work.
Arkeo agents
- How it works
- Each agent owns one recurring job in the software you already use, like your CRM, your inbox or your weekly numbers. We set it up with you, and it reports to your team in Slack or Teams.
- Good for
- Owners of 10 to 50 person businesses who want jobs done end to end, not more tasks to hand out.
- Where it breaks
- Anything your rules keep for a person, like a large discount or an unhappy customer, goes to your team in one message.
Jobs to hand to agents
Agents fit the work that comes back every day or week and crosses more than one program. Keeping your CRM right after each call and email. Keeping the owner's inbox and calendar in order. Pulling the week's numbers into one scorecard each Monday.
Each agent reads what is in front of it, follows your rules, checks its own work, and tells a person when the job needs one. Your team keeps using ChatGPT for its own tasks, and the jobs that come back each week get an owner.
Agents that own whole jobs
Each of these owns one job from start to finish, and your team talks to it in Slack or Teams.
Your inbox, calendar and documents, kept up for you.
- Reads and answers your emailall day
- Books and moves your meetingsas they come
- Writes your docs and deckson request
Your CRM stays right without anyone typing.
- Logs emails and callsas they happen
- Checks off the task when the reply landsright away
- Moves the deal to the right stageper deal
The Monday meeting starts from the same page.
- Pulls your numbers from five toolsMondays at 7 a.m.
- Calls out anything that moved past your set percentageeach week
- Posts a one-page summary to Slackevery Monday
Questions
Is my customers' information safe?
Yes. Agents work in your own accounts and see only what each job needs. Customer details never go into public AI. Agents connect to AI through business-grade connections audited under SOC 2.
What is ChatGPT agent mode?
It was a mode in ChatGPT where ChatGPT did a task for you on its own virtual computer, using websites, code and apps like Gmail. OpenAI's help page now says it is no longer available and points to ChatGPT Work.
What replaced ChatGPT agent mode?
ChatGPT Work, for longer tasks one person hands off, and workspace agents, which teams on Business, Enterprise and Edu plans build, share and put on a schedule.
Can my team keep using ChatGPT?
Yes. Your team can keep using ChatGPT for its own tasks. Agents take the jobs that come back every day and week.
What does it cost?
You get the price on the call, along with the hours it saves your team.