Chatbot or agent?
Recognise your first, and possibly only, AI agent
One question tells you whether the thing you are using is a chatbot or an agent. If you have the Claude or ChatGPT desktop app on your computer, you already have the only one you need. You do not need a team of thirty-seven of them, and the people who build these for a living will tell you the same thing. What is missing is one job and one connection.
Sources opened and quoted on 17 and 22 September 2026. Last checked 2 October 2026.
The question that settles it
Ask yourself one thing about the AI tool you are using right now.
Did anything change outside of the chat?
Text only
You ask it to help you draft an email. You bat it back and forth. Then you copy the output and paste it somewhere yourself.
Nothing moved. You were the hands.
Doing something
You ask it to go into your Gmail and draft a response to something you received. You ask it to make the file. It reaches something outside the chat window.
Now the real question starts: how much of it did it work out for itself?
No, and it is a chatbot. Yes, and it is doing something.
The word gets used two ways. Something that reaches your Gmail on a fixed set of steps you laid out is using tools, but it is following your path. The stricter meaning of agent is about who decides the path.
Anthropic’s engineers draw it against workflows, not against chatbots, and the difference is exactly who is choosing the next step:
“Workflows are systems where LLMs and tools are orchestrated through predefined code paths. Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.”
So it is a scale, not a switch. Text only at one end. Then tools on a path you set. Then something working out its own steps toward a goal you handed it. Do not get precious about which word applies to yours. The first useful thing you build will probably sit in the middle, and it will still save you the afternoon.
You do not need to buy anything. The Claude or ChatGPT desktop app on your computer is the one. On its own it is text only, so the rest of this page is one job to point it at and one connection to turn on.
You do not need a team of them
Thirty-seven of them, each with their own individual coffee orders. You have seen the diagrams. An AI chief of staff, twelve AI employees, each with a name and a job title. A lot of those setups are org chart rather than output: they add complexity without adding useful work.
Anthropic, writing for the engineers who build these systems professionally:
“we recommend finding the simplest solution possible, and only increasing complexity when needed.”
“Start with simple prompts, optimize them with comprehensive evaluation, and add multi-step agentic systems only when simpler solutions fall short.”
So build one. One job, done properly, that you use. That is further than most people with an org chart full of robots.
Pick the job before you pick the tool
This is the step people skip, and it is the step that decides whether the thing is useful. Run your idea past these four. If it fails one, pick a different job.
You already do it by hand
Do not start with something you have never done. You cannot tell whether the output is right if you have no idea what right looks like.
It comes round again
Weekly is a handy rule of thumb, not a rule. A monthly or quarterly job can be well worth it if it is valuable, or if you dread it. The thing to avoid is something so rare that you will have forgotten how your own setup works by the time it comes back.
You can say what a good one looks like
Not necessarily in one sentence. A short list of criteria works, so does one good example and one bad one, so does a checklist you read it against, or simply you reading it before it goes. What you cannot do is leave quality undefined and expect it to land. This is the step most people skip.
Being wrong is cheap
Your first one should not be the thing that goes to a client, moves money, or cannot be undone. Pick something where a bad draft costs you two minutes.
Do not start from a blank page
You know your job better than any AI does. You have just never written it down, so the setup stalls at the first empty box.
So do not write it. Get interviewed instead. Paste this into ChatGPT, Claude, or whatever you use. It asks you six questions and hands you the instructions at the end.
The interview prompt
I want to set up my first AI agent. Interview me before you build anything. Ask me one question at a time and wait for my answer. Work through these six, in order: 1. The job. What is one task I do by hand most weeks? Make me walk you through one real example from start to finish. 2. The trigger. What tells me it is time to do it? 3. The inputs. What do I have to look at to do it, and where does each one actually live? 4. The output. What exists at the end? A draft, a file, a message, a row in a sheet? 5. The finish line. How do I know it was done well? Ask me for one good example and one bad one. 6. The boundary. Which steps may you do on your own, and which have to come back to me before anything leaves? Rules: - One question at a time. Keep the questions short. - If my answer is vague, ask me for a concrete example before you move on. - Do not suggest any tools until you have asked all six. - Do not build anything until I say go. Then give me back four things: - The job written as instructions I can paste into a project or a custom instructions box. - A list of exactly which connections or permissions it needs, and what it can still do for me if I grant none of them. - Anything in this job that touches client, health, financial, or otherwise confidential information, and what that should change about what I connect. - The one step you recommend I keep as a draft rather than let you send. Use my words. Do not tidy my voice into something more corporate.
What comes back is your first version. It will be wrong in places. Fix it the second time you catch yourself correcting the same thing twice.
Turn on one connection, not nine
A connection is what lets the AI reach something outside the chat. It is the part that turns the answer into an action. Anthropic describes it plainly:
“Connectors let Claude access your apps and services, retrieve your data, and take actions within connected services.”
Grant one. The one your chosen job needs. You can always add another next week.
In Claude
These are Anthropic’s own steps, in their own words, as their help pages read on 17 September 2026. Treat every line in this box as what the documentation said that day, not as a permanent fact. What you can switch on depends on your plan, on your workspace administrator, and on your own settings, and products change.
- 1Click the "+" button in the lower left corner of your chat, or type "/" to open the menu.
- 2Hover over "Connectors."
- 3Select "Manage connectors."
- 4Click the "+" button next to Connectors.
- 5Click the connector you want, then Connect or Install, and follow the sign-in prompts.
Gmail, Google Calendar, and Google Drive are “available for all users on Claude and Claude Desktop”. On a Team or Enterprise plan an owner has to switch them on for the organisation first. Anthropic also states: “We do not train our models on your Gmail, Drive, or Calendar connector data”. Google Workspace connectors.
In ChatGPT
OpenAI calls these apps or plugins, not connectors, so that is the word to look for in the menu.
- 1Open Settings, then Plugins. Select the app, or the plugin that includes it.
- 2Select Install plugin if it is shown. That may also start the setup flow.
- 3If prompted, select Connect, or continue the setup flow.
- 4Sign in to the account that already has access to what you want to use.
- 5Review the services and permissions it asks for.
- 6Complete any authorisation the provider asks for.
To change it later, open Settings, select Apps or Plugins, then the connected account. OpenAI states: “App availability depends on the app, your ChatGPT plan, region, workspace, role, model, and the ChatGPT interface you use.” Connecting and managing app accounts in ChatGPT.
Drafting and sending are two different jobs
Do not skip this one. Write down what the agent may do on its own, and what has to come back to you before it leaves the building.
Reading is not harmless. Point an AI at your inbox or your files and it can reach client details, health information, contracts, and things people told you in confidence. If you work in a regulated field, that is a question about what you are permitted to connect, and it is worth answering before you connect it rather than after. Drafting carries its own tail: a draft can be wrong about a person, quote a price you did not agree, or sit in a shared folder where someone assumes it was approved.
So the split is not risky against safe. It is reversible against not. Reading, summarising, organising, and drafting leave you a step where you can still look. Sending, posting, spending, signing, agreeing, and contacting someone new do not. Those can be difficult or costly to reverse, and they land on another person straight away.
For a first setup, put everything in the second list behind your own yes.
- Read the mailbox or folder I pointed it at, and tell me what needs an answer
- Write the reply and leave it as a draft
- Make the file, the summary, or the list
- Send the reply without showing me. Bring it back to me first
- Message anyone I have not already agreed to message. Bring it back to me first
- Spend money, or agree to anything on my behalf. Bring it back to me first
On Claude, the documented default is already on your side. Anthropic’s Google Workspace page, read on 17 September 2026, says the Gmail connector can “Send, reply to, and forward emails from Gmail”, and Anthropic says that “By default, Claude asks for your approval before each of these actions.” A default is a starting position, not a guarantee. On Team and Enterprise plans an administrator can change it, so check what yours is set to rather than assuming. Where it is on, leave it on. It is not friction, it is the thing standing between a helpful draft and an email you cannot unsend.
What I could not check
- The ChatGPT steps are quoted from OpenAI's help centre as it read on 22 September 2026, and rechecked on 2 October 2026. Both help centres describe what the feature is meant to do, and neither is a test of it on your account. Menu names change often, and OpenAI has already moved from connectors to apps and plugins.
- Everything in the Claude box is quoted from Anthropic's help pages as they read on 17 September 2026. It describes what the feature is meant to do. It is not a test of whether it works well on your inbox, and I have not measured that. Plan, workspace administrator settings, your own settings, and future product changes all sit between those pages and what you will see.
- "Building effective agents" is written for engineers building software. I am borrowing its definition and its warning about complexity, both of which stand on their own. The rest of that post is about code you are not writing, and it is not a guide to setting up a connector.
- The scale in section 01 is my framing, not Anthropic's. They draw one line, between workflows and agents. Putting plain chat at one end of it is me, because that is the question people arrive with.
- Whether you are allowed to connect a given mailbox or folder is a question about your obligations, not about the software. I cannot answer it for you and this page does not try to.
The agent is the easy part
Setting one up takes an afternoon. Getting a good one takes something else, and it is not a better model.
An agent with no context is a new hire on day one with no briefing. It does not know what you sell, who you sell it to, how you sound, or what it is not allowed to touch. You can hand it all of that once, in plain files you own, instead of explaining yourself in every new chat.
Build your folder brain is the next thing to read. Four questions, four files, and every agent you make after this one stands on top of them.