All posts
AI Employee#is ChatGPT an AI agent#ChatGPT agent#ChatGPT agent mode

Is ChatGPT an AI agent? What the word actually means

OpenAI now calls ChatGPT an AI agent. Here is what an AI agent is, how a chatbot and an LLM differ from one, what happened to ChatGPT agent mode, which parts of ChatGPT act like an agent today, and what that means for a business choosing one.

Md Shohel· October 16, 2026· 14 min read
Listen to this post0:00 / 15:01
AI narration of this post, read in full

Ask ten people whether ChatGPT is an AI agent and you will get three answers. Some say no, it is a chatbot: you type, it replies, nothing happens until you type again. Some say yes, because it can browse the web and use your apps. And a fair number will tell you it used to have an agent mode, then they went looking for it in August and it was gone.

All three are describing something real. ChatGPT has changed shape several times in under two years, and the word "agent" has been stretched to cover almost anything with a chat box. So this post does two jobs. It pins down what an AI agent is, in terms you can check against any product. Then it walks through ChatGPT as it stands in October 2026 and says which parts meet that bar and which do not.

We should say up front that we make Sanaf, an AI employee that works in Slack and Microsoft Teams. That gives us an interest in how this question gets answered. We have tried to keep the product out of it until the end, and most of what follows is drawn from OpenAI's own documentation, linked as we go, which we read on 16 October 2026.

The short answer

Yes, by OpenAI's own description, and with a condition attached. OpenAI's "Use ChatGPT" guide now opens with the sentence "ChatGPT is an AI agent that you communicate with in natural language." That is a long way from how the product launched in 2022.

The condition is that ChatGPT is several things in one window. When you ask it a question in Chat and it answers, it is behaving like a chatbot sitting on a large language model. When you switch to Work and hand it a task with an outcome, it plans, uses tools, checks its own results and keeps going until there is something to review. That second behavior is what people mean by an agent. Whether ChatGPT counts as an AI agent depends on which of those you are using, and whether anyone is there to start it.

What is an AI agent?

An AI agent is software that is given a goal and works toward it in steps, choosing its own next move. It reads the situation, decides what to do, uses a tool (a search, a file, an app, a browser), looks at what came back and decides again. It stops when the goal is met or when it needs a person.

Anthropic's engineering team draws the line in a way that holds up well. A workflow is a system where language models and tools follow paths a developer set in advance. An agent is one where the model directs its own process and decides which tools to use and how. The test is the loop: does the software decide what happens next, or did somebody write that down beforehand?

Three features show up in almost every serious definition:

  1. A goal larger than one reply. "Draft a reply to this email" is one step. "Clear the shared inbox and flag anything that needs the owner" is a goal.
  2. Tools it can choose to use. Reading a file, searching the web, opening an app, sending a message.
  3. Some autonomy over the steps. It works out the order itself and adjusts when a step fails.

There is a fourth feature that definitions often skip and businesses care about most: knowing when to stop and ask. An agent that can send email on your behalf needs a rule about which emails it sends alone. We wrote more on this in what agentic AI means in plain terms.

AI agent vs chatbot vs LLM

These three words get swapped around constantly, and much of the confusion about ChatGPT comes from that.

Large language model (LLM)ChatbotAI agent
What it isA model that predicts textAn interface for a conversationA system that pursues a goal with tools
What it does with a requestProduces a responseReplies, then waitsPlans steps, acts, checks results, continues
Uses toolsNo, on its ownSometimes, when askedYes, and chooses which ones
Who decides the next stepNobody, it stops at the answerThe person typingThe agent, inside limits set by people
ExampleGPT-6 Luna, Claude, GeminiA website help widget, plain ChatGPT chatChatGPT Work on a task, a coding agent, an AI employee

Is ChatGPT an LLM?

No. ChatGPT is a product built on top of LLMs. OpenAI's current models, such as the GPT-6 family, are the LLMs; ChatGPT is the app that wraps them with memory, files, tools, plugins and the interface you type into. Asking whether ChatGPT is an LLM is a bit like asking whether a car is an engine.

What is the difference between an AI agent and a chatbot?

A chatbot's turn ends when it has replied, and nothing more happens until a person types again. An agent's turn ends when the goal is met or when it reaches a point that needs a person. Plenty of products are both, depending on the mode, which is exactly ChatGPT's situation.

LLM vs AI agent

An LLM is one part of an agent, usually the part that reasons and writes. The agent is the LLM plus the loop, the tools, the memory and the rules. Two agents built on the same model can behave very differently because those other parts differ.

How ChatGPT got here: Operator, agent mode and Work

A short history clears up most of the confusion around searches for "ChatGPT agent".

January 2025: Operator

OpenAI released Operator, a research preview that could use a web browser for you by looking at screenshots and clicking and typing. It started with Pro subscribers in the US, as Wikipedia's Operator entry records.

July 2025: ChatGPT agent

OpenAI folded Operator's browser control and its deep research tool into a single feature inside ChatGPT called ChatGPT agent, usually known as agent mode. You picked it from the tools menu, gave it a task, and it worked on a virtual computer of its own, asking you to take over for sign-ins. Operator itself was retired later that summer.

April 2026: workspace agents

For business plans, OpenAI announced workspace agents that teams could build once, share and run in ChatGPT or in Slack. The Decoder's write-up described them as a research preview from 22 April 2026, with sensitive actions such as sending an email able to require approval.

July and August 2026: agent mode removed

The agent entry started disappearing from ChatGPT in July, and by early August users were seeing a notice that it was no longer available and pointing them to Work. On OpenAI's developer forum, a user who had relied on it complained there had been no deprecation period. OpenAI's support account replied on 7 September that "Work now has a separate cloud browser that can navigate interactive sites and pause for sign-in or takeover on supported accounts."

So if you searched for "how to use ChatGPT agent" and could not find the button, you did nothing wrong. The feature is gone, and its jobs moved into ChatGPT Work and the Browser.

A caveat about sources: OpenAI's help center was behind a bot check when we tried to read it for this post, so the agent mode dates come from the forum thread and press coverage above, while everything about the current product comes from OpenAI's own documentation at learn.chatgpt.com.

Which parts of ChatGPT act like an agent today

Here is how ChatGPT looks in October 2026, read from OpenAI's documentation. Each part is measured against the definition above.

Chat: an assistant with tools

Chat is for a question or a back-and-forth. It can search the web, read a file you upload and draft text. It can call a tool when you ask it to. It does not set itself a goal and work toward it, and it does nothing until you type. Chat on its own is the chatbot half of ChatGPT.

ChatGPT Work: the agent half

Work is where the agent behavior lives now. OpenAI describes it as "a way to delegate real work to ChatGPT" and says it can "plan a task, gather context, use tools, and carry the work through to a result you can review." It can use your files and plugins, run code, build documents and spreadsheets, split tasks across subagents, and in the desktop app use local files and a built-in browser it can click and type in. It asks for confirmation before sensitive actions, and OpenAI's docs say you can "approve important actions" as it goes.

By any of the definitions above, that is an AI agent. It is also, for now, an agent that a person starts. You open Work, describe the outcome and review what comes back.

Scheduled tasks and event triggers

Scheduled tasks let a saved prompt run on a timetable, and on eligible plans they can also fire when something happens in Gmail, Slack or GitHub, such as a new message arriving. This is the part that answers the "who starts it" question. A task that runs every Monday, or whenever a certain email lands, no longer depends on a person remembering to open the app.

For business workspaces there are also Team Tasks, which belong to a team rather than a person and run in the cloud through the team's service account and shared app connections.

@ChatGPT in Slack and Teams

Admins can now set up @ChatGPT in Slack and Microsoft Teams, where people mention it in a channel to summarize, draft or use approved tools. It works through a service account with connections the admin chooses, and OpenAI notes that it "does not inherit a person's file, app, or private-channel access." That is a sensible design, and it is very close to how a shared hire should get access.

Dots: an always-on agent of your own

The newest piece is dots, which OpenAI describes as "an always-on agent that keeps work moving across your tools and projects." A dot has its own computer and browser in the cloud, keeps working when your laptop is off, can be reached in ChatGPT, Slack or Teams, and comes back to you for decisions. Dots are rolling out gradually, so not every account has one yet.

A dot is personal. Its handle carries your name, and by default it responds to you in Slack, although you can tell it to engage with others. That is a design choice worth noticing, because it shapes who the agent works for.

So, is ChatGPT an AI agent?

Put the pieces together and the honest answer is: ChatGPT is a product that contains an AI agent, and in its newest parts it is turning into one you do not have to start by hand.

  • Plain Chat is a chatbot on an LLM, with tools when you ask.
  • Work is an AI agent that a person starts and supervises.
  • Scheduled tasks, event triggers and Team Tasks give that agent a way to start without a person.
  • Dots are an always-on personal agent, still rolling out.

That is why the arguments online never settle. Someone who uses ChatGPT for quick answers is right to call it a chatbot. Someone who has handed Work a three-hour research task is right to call it an agent. They are using different parts of it.

What this means if you are choosing one for a business

For a person, the question "is it an agent?" is mostly curiosity. For a business it decides who ends up doing the work. Before you pay for any AI that calls itself an agent, ChatGPT included, these questions get to the point faster than any label.

  1. Who starts the work? If a person has to open it every time, it saves time on tasks people remember to do. Repetitive admin usually goes undone because nobody remembered.
  2. Who does it work for? A personal agent works for one person and sees what that person can see. A shared one works for the team, and its access belongs to the business. Ask which you are buying, and what happens when the person who set it up leaves.
  3. What will it do without asking? Find where that is set. A good product lets you widen it one kind of action at a time.
  4. Where does the team see its work? If the answer is a separate app, ask how often people will open it.
  5. What is the record? Every action should leave a log that someone who was not there can read and follow.
  6. How is it priced? Per seat suits a tool each person uses. It fits awkwardly on an agent the whole team shares. ChatGPT's own pricing for Work runs per user, with Business at $20 per user per month billed annually or $25 monthly, and Work usage shared with Codex.

ChatGPT answers several of these well, and it keeps improving. For a team whose people mainly need a strong assistant they each drive, a ChatGPT business plan is a reasonable buy, and our ChatGPT comparison says so plainly. We go further into that choice in ChatGPT for business: is it enough?.

Where Sanaf fits

Sanaf is an AI employee: an agent set up as a member of the team, with a job, rather than an assistant each person drives. You hire it once into Slack or Microsoft Teams and give it jobs from a roster of thirty five, such as the inbox, lead follow-up, booking and invoicing. It works inside the apps you already run, through connections you approve one app at a time, and it does each job on its own schedule or when something arrives, so nobody has to remember to start it.

It works for the team rather than one person. Everyone talks to the same hire in the channels they already read, and its access and memory belong to the business. If the work grows, Sanaf acts as the chief of staff and other jobs get agents of their own that report to it.

It asks before anything it cannot undo. You decide what it may do alone, one kind of action at a time, and every job leaves a work log anyone on the team can read. You start with $120 of free credit and no card, then buy credits from $20 when you want them. They never expire, and you pay for the work rather than per seat. The pricing page has the details, and hiring an AI employee shows what the first day looks like.

If you want a capable assistant for each person, ChatGPT is a good one. If you want a job taken off the team's plate and done every day, that is a different purchase, and it is the one we built Sanaf for.

Questions people ask about ChatGPT and AI agents

Is ChatGPT an AI agent?

OpenAI now describes ChatGPT as "an AI agent that you communicate with in natural language." In practice it depends on the mode. Plain Chat behaves like a chatbot that answers and waits. ChatGPT Work behaves like an agent: it plans a task, uses tools, checks results and carries the work to something you can review. Scheduled tasks and dots let it work without a person starting each run.

Is ChatGPT agentic AI?

Parts of it are. A single chat reply is not agentic, because there is no goal being pursued across steps. ChatGPT Work, the built-in browser with Computer Use, and scheduled or event-triggered tasks are agentic, because the system chooses its own steps and tools toward an outcome.

What happened to ChatGPT agent mode?

ChatGPT agent, launched in July 2025, combined Operator's browser control with deep research. It began disappearing from ChatGPT in July 2026, and by early August OpenAI was pointing users to ChatGPT Work instead. Work now has its own cloud browser that can pause for you to sign in or take over.

How do I use ChatGPT as an agent now?

Switch to Work in ChatGPT and describe the outcome you want, with the files or connected apps it should use. For website tasks, ask it to use the Browser. To have it run without you starting it, turn the task into a scheduled task or, on eligible plans, one triggered by a Gmail, Slack or GitHub event.

What is the difference between an AI agent and a chatbot?

A chatbot replies to a message and waits for the next one. An AI agent is given a goal and works toward it in steps, choosing tools, checking results and deciding what to do next until the goal is met or it needs a person.

Is ChatGPT an LLM?

No. ChatGPT is an application built on large language models. The LLMs are OpenAI's models, such as the GPT-6 family. ChatGPT adds the interface, memory, files, tools, plugins and agent features around them.

What is the difference between an LLM and an AI agent?

An LLM takes text in and produces text out. An AI agent uses an LLM as its reasoning part and adds a loop, tools, memory and rules, so it can take actions toward a goal over several steps instead of stopping at one answer.

Can ChatGPT work in Slack and Microsoft Teams?

Yes. Business and Enterprise admins can set up @ChatGPT in Slack and Teams, where people mention it to summarize, draft or use approved tools through a service account. Dots, as they roll out, can also be messaged in Slack and Teams.

Is an AI employee the same as an AI agent?

An AI employee is an AI agent set up as a role on the team. It has a job, connections to the apps that job needs, rules about what it may do alone, memory of the business, a place in the team's chat and a work log. Every AI employee is an agent; most agents are not set up to be employees.

Get the next one by email

We write when we have learned something worth passing on, so nothing arrives unless it is worth your time.

Unsubscribe in one click, any time. Privacy policy