An existing business already has software, knowledge, permissions and ways of making decisions. An intelligent system has to enter that environment. Missing context, unreliable tools or an unclear boundary of authority can undermine an otherwise capable model.
Start with the whole path.
Follow a task from its initial request to a useful outcome. Each transition has a responsibility: retrieve appropriate context, select a route, execute a tool, check the result or ask a person to decide.
Use established components where they fit. Build where the problem requires something different. Their interfaces deserve the same care as the components themselves.
The frontier is in the combination.
Let research change an engineering decision.
Applied research resolves an uncertainty worth acting on. A useful experiment can tell you which part to change, which assumption failed or which path is no longer worth pursuing.
Integration then brings the result into a real workflow—with tests, observability and a way for the organisation to keep improving it.
The Operating knowledge reference study makes one small part of that argument visible: an answer depends on the evidence the system is permitted to use.