Reading the names: model, harness, application and provider
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
A model, an application and the company supplying them can share a name. That makes an apparently simple question—“Which AI are you using?”—surprisingly ambiguous. Think of the model as the engine and the surrounding software as the vehicle. The distinction helps explain why choosing a model is only part of choosing an AI system.
The model is the trained component that interprets information and produces responses. Training means adjusting its internal numerical settings using examples. A model can suggest an action or generate an answer, but it does not, by itself, provide the complete working environment in which a person uses that capability.
A harness is the surrounding software that supplies relevant information, makes tools available and manages repeated steps. A tool is a callable function, such as searching a folder or running a calculation. The harness decides which proposed calls are permitted, carries them out and returns their results. It may also preserve progress and ask for human approval.
The application or workspace is the product the person uses: the conversation window, files, editor and review controls. It may contain a harness and offer a choice of models. A model access service provides the route to a model running on another computer. A provider may develop the model, operate the computer serving it, or do both.
In one setup, Goose supplied the agent environment, DeepSeek supplied the model and OpenRouter supplied access to a hosted model service. Another example used Gemma running locally through Goose. These combinations illustrate separate choices, not a ranking of the products.
“Free,” “open” and “local” answer different questions. Free concerns charges. Open concerns what is available and what you are allowed to do with it. Local concerns where computation runs. A free application can call a paid model, and an application using a local model can still send information to an external tool.
For example, in a clinic referral assistant, the model interprets a request, the harness retrieves permitted records and manages actions, the application lets staff inspect the work, and the access service runs or routes model requests. Improving the model will not repair an outdated referral status or a missing approval step. When comparing two arrangements, identify which layer changed and whether the whole task improved.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy