AI (artificial intelligence) refers to software that learns from data to make decisions without every rule being hand-coded. But the term is an umbrella — several very different technologies live underneath, and marketing language blurs them together.
These three are the most frequently confused. The relationship is hierarchical: deep learning is part of machine learning, which is part of AI.
A large language model (LLM) is a deep-learning model trained to produce and understand text. GPT (Generative Pre-trained Transformer) is OpenAI's LLM family; Claude is Anthropic's; Gemini is Google's. Generative AI (GenAI) covers all models that produce new content — text, images, code, audio. LLMs are the most common form of generative AI in a business context.
These are the three ways to make an LLM work for a company's needs. They are complementary, not alternatives.
An AI agent is a system that not only answers questions but chains decisions and takes actions — opens a ticket, updates a record, calls another system. At its simplest: LLM + tools + memory. Agent architecture is the structure where multiple agents are centrally managed: each has a role, a guardrails layer supervises what they can do, and the whole flow is traceable.
AI in production is never finished — it must be continuously supervised. These terms describe that work.