What is a plugin?
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 plugin is a package that adds capabilities to a larger application. In an agent application, a plugin might bring together several skills and a coordinating procedure. A skill is a reusable set of instructions for a task. The plugin is the container through which such capabilities can be installed or made available together.
Think of a document-preparation plugin. It might include instructions for drafting a report, a template for formatting it and a tool for reading source files. A template is a reusable starting structure. A tool is a software function the application can call. Bundling these parts can make a recurring kind of work easier to set up.
The name is not a universal technical standard. In different products, a plugin may contain code, instructions, connections to external services or some combination. Two products both claiming to support plugins do not necessarily accept the same package format or provide the same controls.
Three ideas are easy to blur. A plugin is a package. A skill is a procedure. A subagent is a separate executing instance assigned part of a task. Installing a plugin does not necessarily create another agent, and adding a skill does not automatically schedule it to run.
A plugin can extend an application without changing its underlying AI model. The model is the trained component that interprets information and generates responses. The plugin may instead change which information, instructions and functions the model can use. This is one reason the same model can behave differently in different working environments.
Additional capability also brings additional responsibility. A plugin that reads files or contacts another service may create new routes through which information travels. Before enabling it, identify what it can access, what it can change and who maintains its instructions or code. A familiar package name is not a substitute for those details.
A clinic-operations plugin could package a referral-review skill, a handoff template and a read-only connection to an approved procedure library. Installing it would make those resources available; it would not mean the agent has authority to contact patients or decide treatment. The organization would need to identify who maintains the procedures and which information the connection exposes. Packaging related capabilities can simplify setup, but the package still inherits the limitations of its instructions, tools and permissions.
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