LLMs in plain English for business owners

What a large language model actually is, what it's reliably good at, and the two failure modes every buyer should understand.

Last reviewed July 13, 2026 aifundamentals

Every AI product you’re being pitched right now — chatbots, “AI employees”, writing tools, agents — has the same engine inside: a large language model, or LLM. Understanding the engine makes you a much harder person to oversell.

The one-paragraph version

An LLM is a system trained on an enormous amount of text until it becomes extremely good at one thing: predicting what text should come next. That sounds trivial. It isn’t. Predicting text well turns out to require absorbing grammar, facts, reasoning patterns, tone, and structure — which is why the same engine can draft an email, summarize a contract, or explain a spreadsheet.

What that means it’s reliably good at

Reading and summarizing. Long email threads, meeting notes, documents — condensed accurately in seconds.
Drafting. First versions of replies, proposals, job posts, descriptions. Humans edit; nobody starts from blank.
Transforming. Same information, different shape: notes → follow-up email, spreadsheet → summary, formal → friendly.
Classifying. “Is this inquiry a sales lead, a support request, or spam?” — the unglamorous skill that powers most useful business automation. See Anatomy of a lead-intake pipeline for it in action.

The two failure modes to respect

1. Confident wrongness. An LLM predicts plausible text, and plausible isn’t always true. Left unsupervised, it will occasionally state something false with total confidence (the industry calls this “hallucination”). The fix is architectural, not hopeful: ground it in your real data, and keep a human approval step wherever an error would cost money or reputation.

2. Inconsistency without instructions. Ask casually, get casual variance. Businesses that get consistent results write down exactly how the model should behave — effectively an employee handbook for the AI. That’s a real discipline; Teaching an AI your way of working covers it.

The buyer’s takeaway

When a vendor says “our AI does X,” translate it: “an LLM, wrapped in some plumbing, does X.” Then ask the two questions that actually matter: what data does it see, and what happens when it’s wrong? Vendors with good answers have built real systems. Vendors who change the subject have built demos.

Unfamiliar terms? The AI & automation glossary translates all of them.