➤ 1. AI (Artificial Intelligence)
The broadest category.
Anything that makes a machine perform tasks that normally require human intelligence falls under AI. This includes:
- Rule-based systems
- Classical machine learning
- Computer vision
- Recommendation engines
- And yes, generative models
Calling everything “AI” is technically correct but almost useless in the engineering conversation. It’s like calling every machine that moves a “transportation device.”
➤ 2. LLM (Large Language Model)
An LLM is a raw technology. Essentially a specialized backend service.
You send it a request (prompt + context), and it returns a response (text, code, reasoning, etc.). That’s it.
By itself, an LLM has:
- No memory of previous interactions (unless you explicitly pass the history)
- No awareness of who or what is calling it
- No ability to take actions, access files, run commands, or interact with the outside world
It’s pure inference. A very powerful prediction engine that knows nothing about the client, the application, or the environment it’s being used in.
Everything beyond that (chat interfaces, tools, memory, agents, coding assistants) is built on top of the LLM, not part of the LLM itself.
➤ 3. AI Coding Agents
This is the current leap. An AI coding agent is a system that gives LLM:
- Tools (terminal, file system, browser, git, test runners, etc.)
- Memory / context management
- Planning and iteration loops
- Permission and safety layers
Examples: Claude Code, Cursor Agent, Devin-style agents, custom setups built with LangGraph, CrewAI, or StrandsAgents.
The key difference:
- An LLM answers questions.
- An AI coding agent tries to complete a goal.
It can explore a repository, write code, run the tests, read the failures, fix the code, and repeat — without you copy-pasting everything back and forth.