Agent vs Chatbot vs Workflow
A chatbot answers, a workflow follows fixed steps you defined, and an agent chooses its own next step based on what it sees — each fits a different kind of problem.
Explain like I'm new to AI
These three get mixed up constantly:
- Chatbot: you talk, it responds. It may use RAG or tools, but it doesn't pursue a multi-step goal on its own.
- Workflow: a fixed sequence of steps a developer wired up in advance (step 1 → 2 → 3). Predictable and reliable.
- Agent: given a goal, it decides each step itself, looping and adapting until done.
The key difference is who decides the control flow — you (workflow) or the model (agent).
- + Predictable, cheap, easy to debug
- + You control every step
- − Can't adapt to surprises
- + Adapts to new information
- + Handles open-ended tasks
- − Less predictable, costlier, harder to debug
Mental model
- Chatbot = a helpful receptionist answering questions.
- Workflow = an assembly line: same steps every time.
- Agent = a capable employee: "get this done" — they figure out the steps.
How it works — the deciding question
Ask: "Do I know the exact steps in advance?"
- Yes, always the same steps → build a workflow. It's cheaper, faster, and predictable.
- No, steps depend on what's discovered along the way → use an agent.
- Just answering questions/conversation → a chatbot (optionally with RAG/tools).
Real-world example
- Chatbot: a support Q&A bot grounded in your docs (RAG).
- Workflow: "summarize this ticket → classify it → route it to a team." Same three steps every time.
- Agent: "investigate why this customer was double-charged and draft a resolution" — the steps depend on what the investigation uncovers.
Technical explanation
Modern practice (echoed by Anthropic's "Building Effective Agents") distinguishes workflows (LLM calls orchestrated along predefined paths) from agents (the LLM dynamically directs its own process and tool use). Workflows offer predictability and lower cost; agents offer flexibility at the cost of latency, spend, and debuggability. The strong recommendation: start with the simplest thing that works — a single prompt, then RAG, then a workflow — and only reach for an agent when the task genuinely needs dynamic decision-making. Many "agents" in production are actually mostly-workflow with one or two agentic steps.
Common mistakes
Building an autonomous agent when a fixed workflow would do. Agents are powerful but less predictable and more expensive — don't add a loop the task doesn't need.
- Calling any tool-using chatbot an "agent" — the test is whether it decides its own steps.
- Using a rigid workflow for a task that truly needs adaptation, then bolting on endless special cases.
When to use it
- Chatbot: conversational Q&A. Workflow: known, repeatable pipelines. Agent: open-ended, adaptive tasks.
When NOT to use it
- Don't use an agent for predictable pipelines, and don't force a workflow onto genuinely open-ended tasks.
Alternatives
- These three (plus plain prompting and RAG) are the palette — most systems mix them.
Quick quiz
Related concepts
- What is an AI Agent? — An AI agent is an LLM that can decide and take actions in a loop, using tools to pursue a goal rather than just replying once.
- The Agent Loop — The agent loop is the cycle of observe, reason, plan, and act that repeats — using tools and results — until the agent reaches its goal.
- How to Design an AI System — Design an AI system in order — job and SLO, shape, knowledge and tools, eval, security, traces, then serving and cost — and do not start with a trophy model.
Last reviewed: 2026-09-04 · Written by ByHeart AI · Reviewed by ByHeart AI