Common evergreen AI terminology and the elements of an AI system.
Model. Input: prompt and context. Output. Tool. Harness. Skill. Follow one familiar task to see how the parts work together.
Progress is saved in this browser.
Common evergreen AI terminology and the elements of an AI system.
Model. Input: prompt and context. Output. Tool. Harness. Skill. Follow one familiar task to see how the parts work together.
Progress is saved in this browser.
The whole system
Our example: turn meeting notes into a weekly project update.
Prompt and context
Processes the input
Ready for your review
ToolReads the project notes
SkillProvides the update procedure
A useful map for generative AI applications. Tools and skills are optional; the model is one part of the system.
AI fundamentals · Model
A trained component that processes input and generates output using patterns learned during training.
Like the engine inside a vehicle: one component of the whole system.
In our weekly update
The model turns meeting notes into a readable project summary.
Keep in mind: Using a model is inference. Supplying notes does not, by itself, retrain it.
AI fundamentals · Input
The instructions and information supplied for a model response. Depending on the system, this can include text, images, audio, and tool results.
The assignment and materials placed on a work desk.
In our weekly update
Your request, the meeting notes, project goals, and a sample update provide the input.
Keep in mind: A file existing in an app does not mean its contents are in the model's input.
AI fundamentals · Prompt
Instructions or questions that guide a response. A prompt can also contain examples and background information.
The brief: what to do, who it is for, and what a useful result looks like.
Try this prompt
Draft a weekly project update for my manager from these meeting notes. Use progress, risks, and next steps. Flag missing information.
Keep in mind: Specify the task, audience, constraints, and format. A prompt can include context.
AI fundamentals · Context
Information available to the model at a particular step, including instructions, relevant history, examples, and retrieved material.
What is on the desk right now, including the brief and supporting documents.
In our weekly update
Meeting notes explain what happened. The deadline and project goals help explain why it matters.
Keep in mind: Prompt and context overlap. Prior chats and files must be supplied or retrieved to be in view.
AI fundamentals · Output
What the model produces: text, structured data, code, media, or a request to use a tool, depending on the model and system.
The draft placed on your desk for review.
In our weekly update
A draft with progress, risks, and next steps, ready to compare with the meeting notes.
Keep in mind: Fluent wording is not proof. Check facts, missing details, and fit to the task.
AI fundamentals · Tool
A function that retrieves information, performs a calculation, or takes an action outside the model.
The equipment available for the job.
In our weekly update
A document tool reads the approved meeting notes and returns their contents.
Keep in mind: The model can request a tool call. Software executes it; the result can become new input.
AI fundamentals · Harness
The software around the model that manages calls and routes tool requests. It may also manage context, state, permissions, and stopping conditions.
The work setup that brings the brief, equipment, and review steps together.
In our weekly update
The harness coordinates access to notes, passes context to the model, runs tool calls, and pauses for draft review.
Keep in mind: Responsibilities vary by product. The harness is software surrounding the model.
AI fundamentals · Skill
A reusable package of task instructions and resources. In the Agent Skills format, instructions live in SKILL.md, with optional scripts and templates.
A playbook for a recurring job.
In our weekly update
A status-update skill packages the procedure, three-section template, and review checklist.
Keep in mind: A skill can guide the use of tools. It does not retrain the model or grant access.
Three different jobs
Think of a kitchen: equipment, coordination, and a recipe.
One skill may use several tools. Writing an instruction does not grant access or enforce a permission.
Follow one run
The harness connects the steps. The model may need more than one turn.
You ask for a weekly update. The harness supplies your request, relevant context, and the status-update skill instructions.
The model requests the meeting notes. The harness checks access and runs the document-reading tool.
The tool returns the notes. The harness adds them to the next model input so the model can use that evidence.
The model produces the update. The application shows the draft, and you check facts and missing information before sharing.
A request to use a tool is model output. The surrounding software executes the operation and returns its result.
Words you will hear
Useful context helps the model answer. It does not guarantee that every statement is correct.
Make it stick
Try an answer, then open the explanation.
Context and output: was the actual deadline provided? Check the draft against the notes and ask it to flag missing facts.
No. Check whether the tool retrieved the contents and the harness included the relevant text in the model input.
A skill. It can supply instructions, examples, and a checklist. The tool provides capabilities; the harness runs the process.
Take it into your work
Choose one AI task you already do. Describe its input, model, output, tools, harness, and any reusable skill.
Then ask: which part needs to improve for a better result?