AI Agent vs Chatbot vs Agentic AI: What Is the Difference?

These terms are used loosely, often by the same vendor for the same product. The distinctions below are the ones that matter in practice: they tell you what a system can do on its own, and therefore what it can get wrong.

The short version

AI agent vs chatbot

The difference is who decides the next step. A chatbot answers your message and stops; you decide what happens next. An agent receives a goal ("find three suppliers, compare prices and draft an email to the cheapest") and works out the steps itself: it searches, reads, calculates and writes, checking the result of each step before choosing the next.

That has two consequences:

• Agents can finish work, not just describe it. The output is a file, a booking, a pull request or a sent message, not advice about how to produce one.
• Agents can make mistakes with effects. A chatbot's wrong answer is text you can ignore; an agent's wrong action may have already happened. That is why agents come with permissions and approval steps.

Many products are both. ChatGPT and Claude started as chatbots and now include agent modes (ChatGPT Work, Claude Cowork) that take multi-step tasks. Customer support "chatbots" increasingly look up orders and issue refunds, which makes them support agents.

AI agent vs agentic AI

An AI agent is a single system: one loop, one set of tools, one goal at a time.

Agentic AI is the wider idea: building software out of agents that plan and act, often several of them coordinating, with people supervising rather than doing each step. When a vendor says "agentic AI platform", it usually means tools to build, run and monitor many agents, such as the frameworks and builders ranked on this site.

In everyday use the two terms overlap, and "agentic" is also used as an adjective for any product with some autonomy ("agentic browser", "agentic IDE").

AI agent vs LLM

The LLM is the engine; the agent is the vehicle. An LLM on its own takes text in and gives text out. An agent adds three things around it: tools (search, code execution, a browser, APIs), a loop that feeds tool results back to the model, and state that tracks the task across steps. The same model can power a chatbot or an agent; what differs is what it is connected to and allowed to do.

AI agent vs workflow automation

In a workflow, a person defines every step and its order; AI may fill in one step, such as classifying an email. In an agent, the model chooses the steps. Workflows are predictable, cheap and easy to audit; agents handle tasks where the steps cannot be known in advance. Many real systems mix them: a fixed workflow with an agent inside the one step that needs judgement. Platforms such as n8n and Dify support both.

AI agent vs skill, prompt and bot

• A prompt is a single instruction to a model.
• A skill (in several agent products) is a packaged set of instructions, and sometimes scripts, that an agent loads when a task calls for it, for example "how to produce our quarterly report". Skills extend an agent; they are not agents themselves.
• A bot is any automated program that acts in a chat or on a platform. Older bots follow fixed rules; an AI agent decides its actions with a model.

How to tell what you are looking at

Ask three questions about any product called an "AI agent":

1. Can it use tools? Search, browse, run code, call APIs, edit files. If not, it is a chatbot.
2. Does it choose its own next step? If every step is predefined, it is a workflow with AI in it.
3. Can it finish the task without you prompting each step? If yes, it is an agent, and you should check what it is allowed to do on its own.

Further reading

• What an AI agent is, with real examples
• How to build an AI agent
• The best AI agents by category
• Definitions of related terms: glossary