Chatbots, AI Agents and Automations: The Plain-English Differences

"Chatbot," "AI agent" and "automation" are often used as if they mean the same thing. They do not, and the differences matter when you are deciding what to buy, what to build, and how much supervision a tool needs.

The short version: an automation follows rules you set, a chatbot holds a conversation, and an AI agent pursues a goal by deciding which steps to take and then taking them. Many real products combine all three.

This guide explains each in plain English, compares how predictable and risky they are, and offers a simple way to choose.

Automations: rules that run themselves

A workflow automation is a set of instructions that runs when something happens. When a form is submitted, add the contact to your records and notify the sales lead. When an invoice is 7 days overdue, send a reminder.

What defines an automation is that you wrote the logic. Given the same input, it does the same thing every time. That makes automations predictable, easy to test and easy to explain. Their limit is that they only handle situations you anticipated. An unexpected input either falls through or follows the wrong path.

Best for: repeatable back-office routines with clear rules, such as lead routing, reminders, record updates and internal handoffs.

Chatbots: conversation, scripted or generated

A chatbot is software you talk to by text or voice. The Consumer Financial Protection Bureau (CFPB), in its June 2023 report on chatbots in consumer finance, describes them as "computer programs that mimic elements of human conversation," and distinguishes two broad kinds.

Rule-based chatbots, in the CFPB's words, "use either decision tree logic or a database of keywords to trigger preset, limited responses." Think of a menu: check order status, book an appointment, see hours.

More advanced chatbots use large language models to generate responses in natural language. They understand a wider range of questions and phrasing, but they can also be wrong in convincing ways. NIST's Generative AI Profile calls this confabulation: "the production of confidently stated but erroneous or false content."

A chatbot's main job is answering. Whether it can also do things, such as book or cancel, depends on what it is connected to. Once it can take actions on its own, it starts to behave like an agent.

Best for: answering common questions from information you control, guiding visitors to the right page or person, and simple, structured requests.

AI agents: goals, decisions and actions

An AI agent is given a goal and tools, and it works out the steps. NIST researchers describe agents as systems that can "automate complex tasks on behalf of users." The OWASP GenAI Security Project describes the key ingredient as agency: "the ability to call functions or interface with other systems via extensions."

In practice, an agent might take a request like "move my Thursday appointment to next week," look up the customer, check the calendar, pick an open slot, confirm with the customer and update the record, choosing each step as it goes.

NIST's AI Risk Management Framework notes that AI systems "are designed to operate with varying levels of autonomy." That range is the important dial. More autonomy means less work for your team and more ways for things to go wrong without anyone noticing. Agents can also be manipulated through the content they read. NIST describes agent hijacking as a type of attack in which someone "inserts malicious instructions into data that may be ingested by an AI agent, causing it to take unintended, harmful actions."

Best for: multi-step tasks that vary from case to case and span several systems, with clear limits on what the agent may do without approval.

Side by side

• Who decides the steps: automation, you do in advance; chatbot, a script or the model decides what to say; agent, the model decides what to do.
• Predictability: automation is highest; rule-based chatbot is high; generative chatbot and agent are lower.
• Flexibility with unexpected requests: automation is lowest; agent is highest.
• Cost of a mistake: depends mostly on what the tool can change. A chatbot that can only talk can mislead; an agent that can issue refunds can spend money.
• Supervision needed: grows with autonomy and with the stakes of the actions allowed.

How to choose

Ask three questions about the task.

1. Are the steps the same every time? If yes, use an automation. It is cheaper to run, easier to test and easier to trust.
2. Is the main need answering questions from information you already have? Use a chatbot, limited to approved content, with an easy path to a person.
3. Does the task require judgment across several steps or systems, and is the volume high enough to justify the oversight? Consider an agent, with tight permissions and human approval for anything costly or hard to undo.

For example, imagine a small physical-therapy clinic. Appointment reminders are a pure automation. Questions about parking, insurance paperwork and what to wear suit a chatbot. Rescheduling, which requires checking the patient's plan of care, the therapist's calendar and the patient's availability, is a candidate for an agent, as long as anything involving clinical advice goes straight to staff.

Combining them well

The strongest setups often mix all three. A voice agent answers the phone and books the appointment; an automation sends the confirmation and reminder; a chatbot on the website handles the same common questions. The guide "Small Business Workflow Automation: Triggers, Actions and Guardrails" covers how to keep the automated pieces safe.

When you evaluate a product, look past the label and ask:

• What actions can it take without a person approving them?
• What systems and data can it access?
• What happens when it does not know the answer?
• How do customers reach a person?
• Can you see a log of what it said and did?

The answers tell you more than whether it is called a bot, an assistant or an agent.

Key takeaways

• Automations follow your rules exactly; chatbots converse; AI agents choose and take steps toward a goal.
• Predictability falls and flexibility rises as you move from automations to agents.
• Generative chatbots and agents can produce confident errors, so limit them to approved information.
• The risk of a tool depends mostly on what it is allowed to do, not what it is called.
• Use the simplest tool that does the job, and add autonomy only with matching oversight.

Sources

• Consumer Financial Protection Bureau, Chatbots in consumer finance (June 2023)
• National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (January 2023)
• National Institute of Standards and Technology, NIST AI 600-1, Generative Artificial Intelligence Profile (July 2024)
• National Institute of Standards and Technology, Technical Blog: Strengthening AI Agent Hijacking Evaluations (January 17, 2025)
• OWASP GenAI Security Project, LLM06:2025 Excessive Agency