AI is like a city

Use the city analogy to understand how the major AI pieces relate without memorizing their technical implementation.

Bottom line

You work through a storefront powered by a model, connected to data and services, using instructions and permissions you control.

01 // The map

Walk through the AI city

02 // Translation

AI concept	City analogy	What to remember
Company	Plot owner	Different companies build and operate different model families
Model	Building	The underlying engine provides the intelligence
Harness	Storefront or workplace	The interface determines how you interact and what actions are available
Context	Office records and filing system	Organized information makes the work more relevant and consistent
Prompt	Work order	Clear instructions define the result, boundaries, inputs, and format
Skill	Employee playbook	Reusable instructions stop you from re-explaining repeat work
Connector	Road or utility connection	A packaged link gives access to an outside service
API	Defined service doorway	Software uses a known contract to request data or actions
MCP	Common adapter shape	AI apps can connect to many systems through a shared standard
Automation	Standing order	A workflow runs on a schedule or trigger

03 // Important limits

A bigger building does not automatically mean a better result for every job.
The same storefront can change the model working behind it.
A connector is not the same thing as the data or service it reaches.
Having a key does not create permission to share what is inside.
Real systems overlap: one product may combine a model, harness, tools, and workflow engine.

04 // Use the map

01
Wrong building?
The model may be too weak, too slow, too expensive, or poorly matched to the task.
02
Wrong storefront?
The harness may not have the files, tools, or working style the task needs.
03
Missing records?
The context may be incomplete, stale, contradictory, or unorganized.
04
Missing connection?
The workflow may need a connector, tool, MCP, API, or manual input.
05
Weak work order?
The prompt or skill may not define done, constraints, or verification clearly enough.

Official sources