Large language model (LLM)
A large language model is software trained on huge collections of text so it can draft, summarise, classify and answer questions in ordinary language by predicting likely wording rather than looking up stored facts.
A large language model is software trained to work with ordinary language. It predicts the next likely words from the text you give it, which is why it can draft, summarise, classify or answer. It isn't a database of company facts. It has no built-in record of your business unless that information is included in the request or retrieved from your systems.
In an operation, the model usually sits behind a specific job. A helpdesk can draft a reply from your knowledge articles. Document processing can pull named fields from a contract. An agent can propose the next step in a process, then stop when a person is needed. You still set the prompt and the allowed sources. You decide which actions can run without approval. Those controls need reviewing as the process changes, or the output drifts.
The same prediction method is where it fails. An LLM can produce a fluent answer that is wrong or invented. It can miss a policy that wasn't in the text it was given. Staff using public chat tools can also paste client data into a service you don't control. Two similar requests can come back differently unless you constrain the format and review the result. Use it for drafting and classification, with a person checking work that affects a customer or a payment.
When it matters
- →Your teams spend hours drafting similar emails, summaries or reports.
- →Incoming tickets or documents need classifying before anyone can act.
- →Staff are pasting client work into public consumer AI tools.
- →You need answers drawn from your own knowledge, not the open internet.
Related terms
Large language model (LLM): common questions
How does a large language model work?
The model is trained on large collections of text. When you send it a prompt, it predicts the next likely words, one after another, until it has a complete reply. That prediction is based on patterns in language, not a lookup of a stored answer. In a business setting you constrain what it can see and what it can do. You supply the relevant document, knowledge article or ticket, and you check the output before it reaches a customer or a system.
What's the difference between an LLM and an AI agent?
A large language model generates and interprets text. An AI agent uses that inside a defined process: it can read a request, apply your rules, take an approved action and record the outcome. The LLM is the language engine. The agent is the job around it, with permissions, stopping points and a human handover. You can use an LLM on its own for drafting. You need an agent when the work has several steps across your systems.
Are large language models accurate?
They can be useful for drafting, summarising and classifying, but they are not a source of guaranteed facts. An LLM can invent a fluent answer when it doesn't have the information. Accuracy improves when you give it your own documents, constrain the format and have a person review anything that reaches a customer or a system. Never treat a chat reply as an approved record. If the output updates a finance system or a customer message, someone accountable should check it first.
Ready to talk?
Book a free, no-obligation discovery call. We'll learn about your business and show you exactly how Wanzo can help — with a bespoke proposal within 48 hours.
No commitment. No sales pressure. Just honest advice.