AI agents vs chatbots: what's the actual difference?

Quick answer

A chatbot responds to what you type. An AI agent executes a task. The distinction is structural: chatbots generate a reply to a single input and stop; agents take actions in external systems, run multiple steps, and keep working until a goal is complete. Most things marketed as "AI" today are chatbots. Agents are a different category entirely.

What chatbots are built for

Chatbots are conversation machines. They take a message in, generate a response, and return it to you. They're well-suited to:

• Q&A and FAQ deflection — answering predictable, repeated questions from a knowledge base
• Scripted flows — guiding a user through a fixed process (returns, onboarding, booking)
• Customer-facing conversation — anything where the goal is a response, not an outcome
• First-line triage — understanding what someone wants before routing to a human

The key constraint is scope: a chatbot's world is the conversation. It doesn't have persistent access to your systems, can't take actions on your behalf across multiple tools, and doesn't remember context between separate sessions unless you've explicitly built that in. Every turn, it's responding to what's in front of it.

What AI agents are built for

Agents are built for tasks, not turns. They're the right tool when completing something requires:

• Action in external systems — reading from a database, writing to a CRM, sending an email, calling an API
• Multiple steps with decisions between them — not a single generation but a loop: act, observe, decide, act again
• Running without being prompted at every step — an agent can be triggered on a schedule or by an event and complete a workflow end-to-end
• Crossing system boundaries — tasks that require pulling data from one place, doing something with it, and pushing a result somewhere else

Finance teams use agents to pull invoices, check payment status, and send reminders. Engineering teams use them to monitor repositories, summarise pull requests, and route issues. Marketing teams use them to research prospects, draft outreach, and log results back to the CRM. These are workflows, not conversations.

The concrete difference

Same request, two tools:

The request: "Find the support tickets opened in the last 48 hours marked high-priority, summarise each one, and assign them to the right team based on category."

A chatbot: It will explain how you could do this, or offer a template you could fill in. Without access to your ticketing system, it can't actually pull the tickets. It might ask you to paste them in — at which point you're doing the work it should be doing.

An AI agent: If it has the right tools and access, it connects to your ticketing system, queries for tickets matching the criteria, reads each one, generates a summary, applies your routing logic, and assigns them. It logs what it did. You get a report when it's finished.

The chatbot ends at the answer. The agent ends at the outcome.

When to use a chatbot

Chatbots are the right tool when:

• The interaction is user-facing and conversational by design
• You need to handle a high volume of simple, repeated queries
• The goal is to reduce human workload on frontline communication
• Scripted flows are enough to handle the majority of cases

Chatbots are also cheaper and simpler to deploy. For FAQ deflection on a support site or a structured onboarding flow, a chatbot is often the better choice — not because agents can't do it, but because agents are heavier infrastructure than the task requires.

When to use an AI agent

Agents are the right tool when:

• The task requires reading or writing to an external system
• Completing it takes more than one step with decisions between steps
• You want it to run automatically, on a trigger or a schedule, without a human initiating each run
• The output is an action or a structured result, not a conversational response

A good test: if a human completing this task would need to open multiple applications, pull data from more than one source, make a judgement call partway through, and then write something back somewhere — that's an agent task, not a chatbot task.

Can chatbots and agents work together?

Yes, and this is increasingly common. A chatbot handles the conversation — it collects context, clarifies intent, and understands what the user wants. When it determines that executing a task is required, it hands off to an agent, which does the work. The agent returns a result; the chatbot presents it back to the user.

This pattern makes sense for customer-facing products where you want the interaction to feel conversational but the work underneath to be automated. The chatbot is the interface. The agent is the engine.

Design your AI agents in Envelope

Envelope turns a plain-language description of what you want an AI agent system to do into a complete design — roles, tools, model assignments, and human review gates. Free to start, no code required.

Start designing →

Frequently asked questions

Are all large language models AI agents?

No. A large language model (LLM) is the underlying technology — a model trained to generate text. A chatbot is a product built on top of an LLM to have conversations. An agent is a system built on top of an LLM that can take actions using tools. The LLM is the engine; chatbots and agents are different vehicle types built around it. Most LLM products you interact with are chatbots, not agents, even if they feel smart.

Do AI agents cost more than chatbots?

Usually, yes — for a few reasons. Agents run more tokens per task (multiple steps, tool call results flowing back into context), call external tools that may have their own API costs, and often use more capable models for the reasoning steps. For high-volume, simple queries, a chatbot is more cost-effective. For complex, multi-step tasks that currently require human time, an agent is usually cheaper than the alternative — not cheaper than a chatbot on a per-query basis.

Will AI agents replace chatbots?

For most use cases, no. Chatbots are the right tool for user-facing conversation at scale; agents are the right tool for executing multi-step workflows. They solve different problems. Where the lines will continue to blur is in products that combine both — a conversational front-end that triggers agent-executed workflows in the background. The chatbot doesn't go away; it becomes the interface for the agent.

What are examples of agents for specific business functions?

Finance: invoice processing, payment status monitoring, expense report drafting. HR: candidate research, interview scheduling, onboarding task automation. Marketing: lead enrichment, content research briefing, campaign reporting. Engineering: PR summarisation, issue triage and routing, test failure diagnosis. Support: ticket classification and routing, knowledge base search and response drafting, escalation detection. Each of these requires action in external systems — the hallmark of an agent task.

Do I need multiple AI agents, or is one enough?

Start with one. A single agent handles more than most people expect when it's given a clear role and the right tools. Multiple agents make sense when a workflow spans genuinely different domains with different tool requirements, when parallel execution would save meaningful time, or when you need a human review gate between major stages. If you can't articulate a specific reason for splitting, don't. See When to use one AI agent vs. many for a decision framework.

Are AI agents secure?

They can be, but security requires deliberate design. The key principles: give each agent only the tools and access it needs for its specific role (least privilege), build human review gates at any step that touches sensitive data or takes consequential action, log what each agent does and why, and scope credentials at the agent level rather than giving a shared admin key to everything. AI agents inherit the security posture of the systems they connect to — poorly scoped access is the most common problem. See AI agent security: what to lock down before you deploy for a full treatment.