The last responsible moment is the point when postponing a decision would eliminate an important option. Before then, waiting can gather information without blocking work; after it, delay blocks progress. In architecture, deciding too early can constrain options, create maintenance costs, and lead to rework when requirements change. section 11.2
To judge whether a decision can wait, ask: “What code or contract will break if I don’t decide this today?” If the answer is nothing, defer and use the time to learn. This is deliberate scheduling, not procrastination. section 11.2.1 section 11.2.3
Table 11.2 Decision reversibility and when to commit
|
Decision
|
Reversibility
|
Last responsible moment
|
|
Which database engine?
|
Hard to reverse after data accumulates
|
Before you write the data access layer
|
|
SQL schema design?
|
Medium—migrations exist but are painful at scale
|
Before you store production data
|
|
API versioning strategy?
|
Hard to reverse after clients depend on it
|
Before external clients integrate
|
|
Internal function naming conventions?
|
Easy to reverse with refactoring tools
|
Anytime
|
|
Deployment region?
|
Medium—data residency laws complicate moves
|
Before you accept user data subject to regulations
|
|
Monolith vs. microservices?
|
Hard to reverse in either direction
|
After you understand domain boundaries (not before)
|
Compare decisions by how hard they are to reverse and what event makes commitment necessary: API versioning matters before external clients integrate, while naming conventions can wait. section 11.2.1
Record pending decisions, their triggers, assumptions, alternatives, and review dates. Revisit them when conditions change; an earlier choice may no longer fit current traffic or constraints. section 11.2.3


