Chatbots, then and now
A chatbot is just software you talk to. The first generation followed scripts and rules; the new generation is powered by large language models that understand free-form language. Learn the difference, how a modern chatbot is built, and when each approach makes sense — right here on the page.
01What a chatbot actually is
If you have ever texted a company and gotten an instant reply, you have already met a chatbot: software you hold a conversation with — by typing or speaking — that tries to answer questions or get something done. There are two broad generations. The first is rule-based (also called scripted): it follows decision trees, menus, and keyword or intent matching. It is predictable and safe, but brittle — step off the script and it fails. The second is LLM-powered: built on a large language model, it understands free-form language and is far more flexible and capable, but it can hallucinate (state something wrong with confidence) and needs guardrails.
- Rule-based bots are deterministic — the same input always gives the same scripted reply.
- LLM-powered bots handle messages they were never explicitly scripted for, but their answers must be checked and constrained.
- Most real production bots are hybrid — rules for the critical flows, an LLM for open conversation, plus a path to a human.
02Same message, two generations
Here is the same customer message — "I need to change my flight" — handled two ways. Flip between a rule-based bot, which only understands what its menu was built for, and an LLM-powered bot, which understands the request in plain language and can act on it.
03Anatomy of a modern chatbot
A capable chatbot is more than "just the LLM." Five parts work together: it understands the user's intent, a dialog manager keeps track of the conversation, the LLM generates language, integrations connect to real systems (look up an order, book a slot), and guardrails keep it safe, on-scope, and able to hand off to a human. Tap each part to see what it does.
Intent (NLU)
Natural-language understanding works out what the user actually wants — their intent — and pulls out the useful details (dates, names, an order number). It turns a free-form sentence like "I need to change my flight" into something the rest of the system can act on.
04Rule-based, LLM-powered, or hybrid?
There is no single "best" kind of chatbot — it depends on the job. Rules give you control and predictability; an LLM gives you flexibility; a hybrid blends both. Switch between them to see when each fits.
Rule-based — scripted & predictable
Follows decision trees, menus, and keyword or intent matching. Because every path is defined in advance, it is predictable and safe — great for narrow, well-defined tasks. The catch: it is brittle and fails the moment a user goes off-script.
LLM-powered — flexible & capable
Built on a large language model, it understands free-form language and can handle requests no one scripted in advance. Far more natural and capable — but it can hallucinate, so its answers and actions need guardrails and checks.
Hybrid — rules for the critical, LLM for the open
Most real production bots combine both: structured intents and rules for critical flows (payments, account changes) and an LLM for open conversation, all wrapped in guardrails with a path to a human. You get flexibility where it helps and control where it matters.