What does production mean in software?
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.
Production is the environment or version of a system that people depend on for ongoing real work. It describes a use context, not how polished an interface looks. A simple application can be in production if it performs an essential daily task; a beautiful demonstration may still be only a prototype.
Consider a booking website used by paying customers. Its production records contain actual reservations, and its messages affect what customers expect to happen. Changing the application therefore concerns more than whether the new screen works. Existing bookings, permissions and unfinished transactions must continue to make sense.
A development environment is where changes are created. A test environment is where behavior can be checked without the same consequences as live use. Teams may use different names or combine environments, but the underlying distinction remains: experimenting on a copy is different from changing the service people currently rely on.
Production responsibilities include maintaining availability, protecting information and responding to failures. Monitoring means repeatedly checking whether the service is operating as expected. An incident is a problem that disrupts or threatens that operation. Someone must have the information and authority needed to investigate and restore the service.
For AI-enabled products, changes can occur beyond the application’s own code. A model provider may update the trained component generating responses. Instructions or source documents may change. These alterations can affect output even when the interface appears identical. The deployed arrangement needs to be understood as a combination of components and settings.
Changes to a live service must preserve customer data and continuity. That is a different standard from making a disposable prototype for feedback. A successful experiment does not automatically establish the ability to maintain a dependable service.
Once a clinic relies on a follow-up queue to organize daily work, that queue is part of production even if its screen is simple. Staff need to know when it was updated, what to do if records stop arriving and where unfinished tasks remain visible during an outage. A model or software change should not silently alter the meaning of “overdue.” Being in production describes real dependence, not proof of quality. The organization must be able to operate, investigate and recover the service.
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