🗓️ 19092026 2037

AI OVERVIEW

AI applications become useful when a language model can use context and act through controlled capabilities. This collection explains the layers around that idea, not how models are trained.

Start here​

An AI application has four separate jobs:

  • Model — turns a request and context into a response or a proposed tool call.
  • Application — supplies context, runs tools, and enforces permissions.
  • Workflow — decides whether one pass is enough or work needs multiple steps.
  • Integration — connects the application to external systems such as files, databases, and APIs.

The model can propose an action. The application must decide whether to allow and execute it. llm_tool_use explains this boundary.

Two learning paths​

Build an agent​

Connect external systems​

How the ideas fit​

user request
|
application / agent workflow
|-- model uses [[llm_tool_use]]
|-- optional memory and review loops
`-- optional MCP client
`-- MCP server
`-- database, files, API, or service

MCP is an integration standard, not an agent framework. It tells an application how to discover and communicate with a capability provider. It does not decide what the model should do or grant permission for an action.

References​