AI Agent vs Chatbot: What's the Difference?

An AI agent can take action across your business systems (book an appointment, issue a refund, update a record), while a chatbot can only answer questions inside a chat window based on scripts or a knowledge base. The short version: chatbots talk, AI agents do. If your tool can't touch your calendar, CRM, or payment system without a human finishing the job, it's a chatbot, not an agent.

Key takeaways

• A chatbot follows pre-programmed conversational paths and answers basic questions; an AI agent connects to company data and takes independent action across workflows, according to Salesforce.
• AI-agent-resolved conversations typically cost $0.30 to $2.00 each, versus $5 to $15 for a human-handled support interaction, per Gartner estimates cited by Quickchat.
• Chatbots are faster at scale: one benchmark from Redis found a chatbot workload sustaining up to 6.4 queries per second, versus 1.2 to 2.6 queries per second for a multi-step reasoning agent.
• Intercom Fin charges $0.99 per resolution and Quickchat AI charges $0.50 per resolution, two real published 2026 pricing examples.
• Most businesses don't need to choose one or the other. They need a chatbot for simple FAQ deflection and an AI agent for anything that requires touching a system of record.

What is a chatbot, exactly?

A chatbot is a conversational program that follows preset rules, decision trees, or a static knowledge base to answer questions, and it generally can't take action outside the conversation itself. Cognigy puts it simply: a chatbot talks to your users to answer questions or provide content. It's good at answering "what are your hours" or "where's my order," but if the answer requires changing something in another system, a chatbot either can't do it or hands the person off to a human.

There are two common flavors. The first is rule-based: if the customer types X, show response Y. The second is a retrieval-based chatbot layered on an LLM, which sounds more natural and can pull from a knowledge base, but according to Redis's framing, it still "takes user input and returns text without directly taking external actions." Neither flavor plans multiple steps or reasons about what to do next. They respond.

What is an AI agent, exactly?

An AI agent is a system that can independently plan and take multi-step actions, like booking an appointment or issuing a refund, by connecting directly to your business tools and data rather than just chatting about them. Salesforce describes AI agents as "autonomous systems that connect directly to company data to reason through complex tasks and take independent action across workflows."

That's the practical difference in one line, and Slack's framing nails it just as tightly: "Chatbots respond. AI agents resolve." An AI agent doesn't just tell a customer their return qualifies for a refund, it processes the refund. It doesn't just tell a lead the earliest appointment slot, it books it and sends the confirmation. That requires tool access (calendar, CRM, payment processor, inventory system) and a reasoning loop that decides what step comes next based on what happened in the last step.

The one-line test: if the system can only tell you what to do next, it's a chatbot. If it can go do it, it's an agent.

What's the actual difference between an AI agent and a chatbot?

The difference comes down to whether the system can act on your data or only talk about it. Here's a side-by-side comparison built for a quick gut-check, not a marketing brochure:

Dimension	Chatbot	AI agent
Core job	Answer questions	Complete tasks
Decision style	Pre-set rules or scripted flows	Reasons through steps, adapts based on results
System access	Usually read-only, a knowledge base or FAQ doc	Read/write access to CRM, calendar, payment, inventory
Handles multi-step requests	Rarely, escalates to a human	Yes, that's the point
Speed at scale	Faster, up to 6.4 queries/sec in one Redis benchmark	Slower per query, 1.2-2.6 queries/sec in the same benchmark, due to multi-step reasoning
Cost per interaction	Often bundled into a flat SaaS seat fee	$0.30-$2.00 per resolved conversation, per Quickchat's 2026 vendor pricing survey
Best fit	High-volume, narrow-scope, well-defined questions	Anything requiring an action outside the chat window

Cognigy's guidance on when to use which is worth repeating directly: the best chatbot use cases are "those with narrow scope and well-defined parameters." Once the request needs judgment, data lookups across systems, or an action taken on someone's behalf, you've stepped into agent territory.

How much does each one actually cost?

Chatbots are usually cheaper per seat but limited in what they solve, while AI agents cost more per resolved conversation but can close out the entire task instead of just deflecting it. ChatBot.com, for example, prices its plans starting at $19/month per seat, with a set number of AI resolutions included (10 on its Essential plan, 200 on Growth, 2,000+ on Enterprise), and $0.99 for each additional resolution beyond that.

On the agent side, published 2026 per-resolution prices include Intercom Fin at $0.99 and Quickchat AI at $0.50, according to Quickchat's own market comparison. Gartner's estimate puts the cost of a human-handled support interaction at $5 to $15 depending on the channel and complexity, which is the number that makes the agent math work even at $1-2 per resolution.

Worked example (illustrative, not a client case): Say a business handles 1,000 support conversations a month, and today a human handles all of them at an average cost of $10 each (mid-point of Gartner's $5-$15 range). That's $10,000/month. If an AI agent resolves 70% of those conversations at $0.99 each (Intercom Fin's rate) and the remaining 30% still need a human at $10 each, the math looks like this:

• AI-resolved: 700 x $0.99 = $693
• Human-escalated: 300 x $10 = $3,000
• Total: $3,693/month, versus $10,000/month fully human

That's a rough 63% reduction in handling cost in this example, before counting the value of faster response times or off-hours coverage. Actual resolution rates and human cost per ticket will vary by business and complexity, so treat this as a model to plug your own numbers into, not a guarantee.

Is a chatbot or an AI agent faster?

A chatbot is faster per query, but an AI agent gets more done per interaction because it completes multi-step work instead of just answering once. A throughput benchmark from Redis found a standard chatbot workload sustaining up to 6.4 queries per second, while a ReAct-style agent (the common "reason, then act" agent pattern) sustained only 1.2 to 2.6 queries per second in the same test setup. Redis notes plainly that the multi-step reasoning reduced throughput.

That tradeoff matters for capacity planning. If you're running a high-volume, simple-question channel (order status, store hours), a chatbot's raw speed wins. If you're running a channel where each conversation needs to end in a booked appointment, a processed refund, or an updated record, the extra latency of an agent is the cost of actually finishing the job instead of punting it to a human.

The Resolve Test: a 3-question framework for choosing between them

Use these three questions to figure out which one you actually need, not which one is trendier:

1. Does answering this question require touching another system? If yes (booking, billing, inventory, scheduling), you need an agent. If no, a chatbot can handle it.
2. Does the request require more than one step to complete? Checking an order status is one step. Rescheduling an appointment, checking availability, and confirming it is three. Multi-step means agent.
3. What happens if it gets it wrong? Low-stakes FAQ misses are annoying. A billing agent making an unauthorized refund is a real problem. Higher stakes generally call for tighter guardrails and human-in-the-loop review, agent or not.

If you answer "yes, multi-step, and touches a real system" to two or more of these, you're not looking for a better chatbot script. You need an agent connected to your actual tools.

What mistakes do businesses make when choosing between them?

The most common mistake is buying a chatbot to solve a problem that actually requires action, then wondering why resolution rates barely move. DevRev's blog describes this pattern directly: "You deflected some FAQs and cut a little handle time, but the hard work still landed with your agents." A chatbot that can quote your return policy but can't process the return just moves the bottleneck, it doesn't remove it.

The second mistake runs the other way: deploying a full reasoning agent for simple, high-volume FAQ traffic where a scripted chatbot would be faster and cheaper per interaction. Not every conversation needs a reasoning loop. Match the tool to the task, not the other way around.

A third mistake is assuming "AI agent" is a single product you buy off a shelf. In practice it's closer to a system you design around your specific workflows, tools, and edge cases. This is exactly the kind of system we design, install, and manage for businesses, because the hard part isn't the model, it's wiring it correctly into your calendar, CRM, and phone system so it can actually finish the job.

Which one should your business actually use?

Most businesses end up needing both, not one instead of the other. A chatbot handles the high-volume, low-stakes questions cheaply and fast. An AI agent handles the requests that need to end in a completed action, like a booked slot, a processed payment, or a callback scheduled.

A good real-world version of an agent doing this: an AI voice agent that answers a missed call, checks calendar availability, books the appointment, and sends a confirmation text, all without a human touching it. That's not a chatbot with a phone number attached, it's a full agent loop: it reasons about availability, takes the booking action, and confirms it. A related pattern is recovering abandoned bookings automatically, which only works because the system can actually act on the calendar, not just describe what a person should do next.

If you're evaluating vendors, ask two blunt questions: what systems can it write to (not just read from), and what happens when it's wrong. Vendors selling "AI agents" that can only search a help doc and hand off to a human are selling a chatbot with better marketing.

If your business is losing bookings, refunds, or leads to the gap between "answered the question" and "finished the task," get in touch. Automation Atlas designs, installs, and manages custom AI agents that connect to your actual calendar, CRM, and phone system, so the conversation ends with something done, not just something said.