An architectural seam is a deliberate boundary that lets part of a system vary, be replaced, extended, tested, reconfigured, or modified without widespread disruption. Its value lies in containing change while preserving useful flexibility; it is not an excuse to abstract every component. section 4.4.2 section 4.7
Design seams where change is likely, while keeping their cost proportionate:
- Isolate volatile dependencies behind stable interfaces, such as a façade or adapter around a changing third-party API. section 4.4.2
- Keep contracts, APIs, message formats, and events stable enough for consumers to survive internal replacement. section 4.7
- Use queues or messaging selectively to separate timing and lifecycle, beginning with a narrow need such as buffering or temporary-failure resilience. section 4.4.2
- Avoid seams everywhere: unnecessary indirection adds cost, so prioritize expensive lock-in, likely change, and uncertain decisions. section 4.4.2 section 4.7
Incremental replacement
A Strangler Fig approach routes selected capabilities from a legacy implementation to a new one, gradually shifting traffic until the old capability can be retired. Routing points, APIs, integration boundaries, or database ownership boundaries provide the seams; coexistence limits the risk of a big-bang rewrite. section 4.4.3
A seam is strongest when replacement can happen in slices and when the boundary absorbs change without forcing a system-wide rewrite. This makes seams practical instruments for feedback, delayed commitments, incremental modernization, and controlled evolution. section 4.10 section 4.7


