Models, tools, agents: the words that explain modern AI
Most confusion about AI comes from mixing up a few layers.
Once you can name them, you can tell why a result was great or useless, and what to change.
The model
A model is a prediction engine trained on enormous amounts of text, images, audio, or video.
You give it input, and it produces the most useful continuation it can. What it knows comes from two places: its training data, which stops at a cutoff date, and whatever you put in front of it right now.
The context window
That second source is the context window, the model's working memory for a single conversation or task.
It is measured in tokens, small chunks of text. In English a token is roughly three quarters of a word. Several current models accept around a million tokens. A bigger window lets you hand over more material, but quality still depends on what you choose to put in it.
Tools
On its own, a model can only produce output.
Tools let it act: search the web, read and write files, run code, drive a browser, check a calendar. The Model Context Protocol (MCP) is the open standard most AI apps now use to plug tools in, so one integration works across many assistants.
The agent
An agent is a model working in a loop.
It reads the goal, chooses a tool, looks at the result, decides the next step, and repeats until the job is done or it gets stuck. The same model can feel ordinary in a chat box and remarkable as an agent, because the loop lets it check its own work.
Skills
Skills are packaged know-how: a folder with instructions, and sometimes scripts or templates, that an agent loads only when a task calls for it.
Tools give an agent hands. Skills teach it how your work is done.
Diagnose any bad result
| Symptom | Usually the layer | Try this |
|---|---|---|
| Confidently wrong about recent events | Model (training cutoff) | Turn on web search or paste the source |
| Generic, could be about anyone | Context | Add your audience, examples, and standards |
| "I can't access that" | Tools | Connect the app through a connector or MCP server |
| Stops halfway or needs you at every step | Agent loop | Use an agent mode and define what done looks like |
| Right once, wrong the next time | Skills | Write the procedure down once as a skill |
Pick one task you repeat every week. Write three lines: the context it needs, the tools it touches, and what finished looks like. That is a working spec for your first agent.
Resources
Every link this breakdown leans on, in one place.