Model, context, skill, tool, agent, squad
Six words look like synonyms when you leave regular chat. Separating them keeps you from asking the model for something that should live somewhere else.
The first stumble after regular chat is calling everything an agent (a program set up to carry out tasks). You switch the AI model (a program that reads requests and generates answers) and expect it to know the company. You write a long instruction and expect it to become a tool (a way for the model to act outside the conversation). You connect a tool and expect it to fill in what the model does not know about the company. Much frustration comes from that confusion.
Before building an automation, I separate six parts. Each one solves a different need.
Model
The model is what reasons. It reads the question, combines patterns, writes, and chooses what to do next. When I switch models, I also change the type and quality of reasoning. One may be more careful, follow a long sequence, or use tools with more discipline.
I switch models when a task needs another kind of reasoning. That can mean difficult architecture, a careful review, or writing with rules. The switch does not teach the model about me. Even a better model does not know my projects unless I give it context.
Context
Context is information the system can use without me explaining everything again. The whole conversation is not good for this. It fades and collects details unrelated to the task. Useful context lives in files: decisions, customer profiles, product rules, glossaries, and histories someone can open again.
I need context when an answer depends on me, the company, or a project. Without it, a proposal, sales plan, or review can become generic. A better model also gives generic work when it receives generic information.
Skill
A skill is a work guide the agent recognizes and follows. It records a way of working, not memories. If I check risks, blockers, and minimum tests every time I review a plan, I can save those steps in a skill. Then I do not have to explain the same guide again.
I use a skill when a task repeats with similar steps. A good skill says how to act, what to check, and when to stop. If information is missing, a skill does not solve the problem. It only helps follow a guide better.
Tool
A tool or MCP (a way for the model to act outside the conversation) can read a database, search the web, open a document, or use another application. Without one, the model only talks. It can write a perfect plan without changing anything.
I use a tool when a task needs something outside the conversation to be checked or changed. For example, check a calendar, look up a record, create a card, or read a file that is not in the context. The tool does not think for you. It gives the model a way to act.
Agent
The role of an agent brings together context, tools, and a goal. You name it when you want to call it again: PR reviewer (a request to review and include code changes), researcher, analyst, or editor. The name saves you from building the same setup again.
An agent makes sense when repeated work needs the same approach and access. If everything changes with every request, a conversation with good instructions may be enough. An agent starts to help when its role is stable.
Squad
A squad brings different agents together for one task. One researches, another implements, and another reviews. This keeps different responsibilities separate.
I only build one when a task has independent parts or needs roles that challenge each other. For a small task, a squad only adds coordination work. It is not the first step.
Almost every diagnosis ends with this question: what do I need now? I can give context, follow a guide, use a tool, or call a specialized role. This avoids asking the model for something that should be written in a file.
To start on the right step, write the context file first.
Where it breaks
It is easy to move up too soon. Almost everyone who wants to build a squad has not written a context file yet. They create several agents with different names, and all return different versions of the same generic answer.
The cost appears when coordination comes before useful content. Without context, a skill becomes an empty guide. A tool becomes a button without judgment. An agent becomes only a personality, and a squad becomes a meeting. Write the foundation before asking the system to work as a team.