The sketch on the home page is the shortest way I have to say this.
A model is the engine. A harness is the vehicle around it: the tools, the files, the rules, and the checks. The question on the sketch is who’s driving.
The cases on this site are written as three stages.
You drive. You ask, you steer, and you check the result. The model answers. You stay in the seat the whole time. On a case, that is the Idea and the How.
The car drives you. You set the destination and the rules, then a harness keeps the model on the road. It can use tools, follow a file of instructions, and stop where it should stop. You still approve the turns that matter. That is Build it.
The car works for you. A scheduled or always-on agent does a defined job and reports back. You wrote the job, and you can shut it off. That is the agent file: one AGENTS.md, with thin pointers for the tools that look for their own filename.
Same car. The only thing that changes is who has the wheel. AI 101 walks through model, harness, and agent in that order, and every use case is one trip along that road.
— Cases4AI, powered by AIRVX