AI agents
What is the difference between a chatbot and an AI agent?
A chatbot is a single turn: you ask, it answers, done. An agent runs a loop, controlling how the work unfolds rather than following a fixed script.
An AI agent is a language model that has been given a goal and the means to pursue it. Instead of writing a single reply, it decides what to do next, takes an action, looks at what came back, and decides again. It keeps going until it judges the goal met or hands control back to you. As one comparison puts it, chatbots handle conversations and agents handle work.[1]
The cleanest way to tell them apart is to ask where the loop is.
A chatbot is a single turn. You ask, the model answers, and that is the end of the exchange. Nothing happens in the world unless you go and do it. Ask a chatbot to fix a failing test and it describes the fix in prose, for you to type in yourself.
An agent runs a loop. It is handed a goal, then it reasons about a step, acts, observes the result, and repeats. Ask an agent the same question and it opens the file, edits the line, reruns the test, and reports back whether it passed — no copying and pasting required. OpenAI draws the line at exactly this point: a system is an agent when the model itself controls how the work unfolds, rather than following a fixed script.[2]
That shift in control comes with a cost. A chatbot’s failure mode is a wrong sentence: annoying, cheap to catch, easy to ignore. An agent’s failure mode is a wrong action — a file edited incorrectly, a message sent to the wrong person, a delete that runs before anyone reviews it. Which tools an agent is handed, and what limits sit around them, end up mattering as much as the model behind the wheel. See the three ingredients and rules and guardrails.
Most products people already use sit on the chatbot side of that line: search summaries, writing assistants, support widgets that draft a reply for a human to send. The agent side is newer and still being worked out — coding assistants that run commands themselves, browser agents that click through a checkout, research agents that chase down sources without being told exactly where to look.
Common agent frameworks
- OpenAI Agents SDK ↗
Anthropic-independent toolkit and AgentKit for building agents, with a widely-read practical guide.
- Anthropic — Building Effective Agents ↗
The reference essay on when to use a workflow versus an agent, with patterns.
- LangGraph ↗
Graph-based framework for stateful, multi-step agents from the LangChain team.
- CrewAI ↗
Framework for coordinating multiple role-playing agents on a shared task.