Large Language Model (LLM)

tl;drA large language model (LLM) is an AI model trained on vast amounts of text that can understand and generate natural language, answer questions, summarize documents, and reason through multi-step instructions.

Claude and GPT-4 are examples. LLMs are the technology that makes AI agents in accounting possible: they're what lets an agent read a messy invoice, understand an ambiguous instruction, or explain a variance in plain English instead of requiring a rigid, pre-defined input format.

An LLM alone isn't an agent. The model provides the reasoning and language understanding; an agent wraps that reasoning in the ability to take action: call the right systems, check the data, apply the company's rules, and complete a task end to end. In accounting specifically, an LLM can read and interpret a bill, but it takes an agent, built on top of it, with access to the ledger, the approval policy, and the audit trail, to actually process that bill correctly and safely.

The distinction matters when evaluating a finance tool that claims to use AI. Understanding language is a small part of the job. Executing correctly, inside the right controls, is the harder and more important part.

Back to the glossary