Chatbot vs Workflow vs Agent
I’ve been deep diving on AI Agents and realized the difference between chatbots, workflows, and agents. Folks tend to use the word “agent” when building an AI system, when they might mean one of the other two. This matters because the loops, as we’ll see, have different shapes.
AI Chatbot
A chatbot is like a question and answer session with an AI model. You give it a prompt/question and it returns an answer to you. That’s it.
You may want to fetch some data from the user as additional information to decorate the prompt, but it’s still just getting an answer from the model and returning it.
AI Workflow
If a chatbot was simply returning the model’s answer to the user, a workflow performs a finite and predetermined set of additional actions based on the model’s response. This may or may not include subsequent calls to the model.
For example, if we’re building a meeting summarization service, if the user supplies raw meeting audio, the model could extract a text transcript, then a second call to the model could summarize into key points.
The user didn’t have to provide any guidance. We know the steps we want to take and that workflow is encoded into the system.
AI Agent
If a workflow was executing a series of predetermined actions based on the model’s output, an AI Agent executes a series of self-determined actions. As in, the model is free to execute actions that it deems necessary to satisfy the user’s question/prompt.
Since the agent determines its own steps, we require the user to provide guidance on what they want done. So for a meeting summarization service, the user could now say “here’s raw meeting audio, give me a summary and key points.” The agent would then determine that it must transcribe the audio first, then loop on itself to summarize and extract the key points from the transcript.
The beauty here is that the agent can now support dynamic workflows (with many combinations of actions) without us hardcoding them.
