From Chatbot to Agent: What Actually Changes
"Agent" has become one of the most used words in AI. Browsers call themselves agents, coding tools call themselves agents, and some products call a chatbot with a new button an agent. It is hard to tell what the word means.
The idea underneath is simple. A chatbot answers you. An agent does things for you: it takes a goal, decides on steps, uses tools to act, checks what happened, and keeps going until the goal is done or it gets stuck. This lesson shows exactly what has to be added to a chatbot to make that possible.
What You'll Learn
- The difference between a chatbot and an agent in one sentence
- The three ingredients that turn a language model into an agent
- A simple scale from "not an agent" to "fully autonomous"
- Why the same model can power both a chatbot and an agent
The one-sentence difference
A chatbot takes your message and writes one reply. Then it waits for you.
An agent takes a goal and works through several steps on its own, using tools, until the goal is reached.
A chatbot talks about the task. An agent does the task.
| Criteria | Chatbot | Agent |
|---|---|---|
| You give it | A message | A goal |
| It gives back | One reply | A finished task, after many steps |
| Who decides the next step | You | The agent |
| Can it act in the world | No, it only writes text | Yes, through tools |
| Example | "Explain how to book a flight" | "Find and hold the cheapest flight on Friday" |
Chatbot
- You give it
- A message
- It gives back
- One reply
- Who decides the next step
- You
- Can it act in the world
- No, it only writes text
- Example
- "Explain how to book a flight"
Agent
- You give it
- A goal
- It gives back
- A finished task, after many steps
- Who decides the next step
- The agent
- Can it act in the world
- Yes, through tools
- Example
- "Find and hold the cheapest flight on Friday"
Three ingredients
The "brain" of almost every agent today is a large language model, the same kind of model behind ChatGPT, Claude, or Gemini. On its own, that model can only read text and write text. Three things are added around it.
1. A goal. Instead of a single question, the agent gets an outcome to reach, such as "summarize every unread email from my manager and draft replies."
2. Tools. A tool is anything the agent can use to act outside the chat: search the web, open a web page, read a file, run code, send an email, add a calendar event. The model cannot do these things itself. It asks for a tool, and the software around it runs the tool and returns the result. Lesson 3 covers this in detail.
3. A loop. After each action, the agent looks at the result and decides what to do next. It repeats until it thinks the goal is done. This loop is the heart of every agent, and lesson 2 walks through one step by step.
- Language modelReads and writes text
- + GoalAn outcome to reach
- + ToolsWays to act outside the chat
- + LoopAct, check, decide again
Some agents also have memory that lasts between sessions, which lesson 4 covers. But goal, tools, and loop are the core.
The same model, two different jobs
This surprises many people: the model inside an agent is usually the same model you chat with. What changes is the software around it.
When you chat, the software sends your message to the model and shows you the reply. When it runs as an agent, the software sends the goal, a list of available tools, and the results so far, then asks the model: "What should happen next?" The model's answer might be a tool request instead of a reply for you. The software runs the tool, adds the result, and asks again.
So when a company launches "an agent," the big change is usually in this wrapper software, the tools it connects to, and how much freedom it gives the model. If you want to understand the model itself, How LLMs Actually Work covers it.
A scale, not a switch
"Agent or not" is less useful than "how much does it do on its own?" Think of it as a scale:
- Chatbot. Answers questions. No tools.
- Chatbot with tools. Can search the web or read a file when you ask, then replies. One step at a time, you stay in control.
- Workflow. A person designed a fixed series of steps, and AI fills in parts of it. For example: when a form arrives, AI summarizes it and the summary is emailed to a team. The path never changes. Tools like Zapier, Make, and n8n often work this way.
- Agent. The AI decides which steps to take and in what order, based on what it finds.
- Autonomous agent. Runs for a long time, handles many steps and surprises, and checks in with a person rarely.
- AI systems that use tools
- Workflows: a person fixes the steps, AI fills in parts
- Agents: the AI chooses the steps as it goes
The difference between a workflow and an agent matters. A workflow is predictable because a person chose the path. An agent is flexible because it chooses its own path, which also means it can choose badly. Many tasks work better as workflows. If the steps are always the same, you do not need an agent. AI Automation Fundamentals covers building workflows without code.
Key Takeaways
- A chatbot writes one reply to a message. An agent works toward a goal over many steps, using tools.
- Agents add three things to a language model: a goal, tools, and a loop.
- The model inside an agent is usually the same model you chat with. The wrapper software and tools make it an agent.
- Think of agents on a scale: chatbot, chatbot with tools, workflow, agent, autonomous agent.
- Workflows follow a fixed path set by a person. Agents choose their own path, which makes them flexible and less predictable.