A large language model is the engine behind tools like Claude and ChatGPT: it reasons over the context you give it rather than looking up facts in a database. It learned patterns in language at massive scale, and most professional work is language work: describing problems, writing instructions, drafting replies. The model plus context plus tools becomes the working agent: in Claude Code, the LLM reads your prompt, looks at files, decides steps, and writes back.
Picture a very well-read assistant across the desk from you. They’ve read manuals, emails, contracts, and a lot of messy human writing. Hand them a note and they read what’s in front of them, recall everything similar they’ve seen, and write the most sensible next thing.
Why you care
Most frustration comes from treating an LLM like a smarter Google. Search finds pages; an LLM generates an answer from patterns, instructions, and context, so it can write smoothly while guessing. If you want it grounded, give it the right material or connect it to RAG. The context window is the desk: leave the important file off it and quality drops before the model even starts.