Abstraction creates a stable boundary while hiding implementation detail, but it also adds complexity. Its value depends on evidence, consequence, and reversibility rather than imagined future variation. section 3.1
Judge an abstraction against the realities of delivery and operation:
- Add it when real variation exists, changing the boundary later would be expensive, or delaying it would make evolution harder. section 3.1
- Defer it when complexity addresses only hypothetical scenarios; preserve a seam or decision record instead of building the full capability. section 3.1
- Prefer patterns the team can operate, debug, and evolve without external heroes. section 3.1
A narrow workflow becomes an architecture exercise
A three-month internal workflow project needs one workflow, one department, and a few known integrations. Splitting it into services, adding messaging, and designing for speculative scale turns abstraction into perfection drift. A simpler boundary is justified until concrete variation or costly reversal provides stronger evidence. section 3.1
Table 3.2 Responding to perfection drift
|
Signal
|
Diagnostic
question
|
Better
response
|
|
Over-engineering
early
|
What
current problem does this complexity solve?
|
Remove or
defer the part that solves only a hypothetical problem. Keep only the
structure needed for the known problem.
|
|
Designing
for hypothetical future scenarios
|
What
evidence suggests this future is likely soon?
|
Preserve a
seam or decision record but avoid building the full future capability until
the signal becomes real.
|
|
Building
abstractions “just in case”
|
How many
real variations exist today?
|
Wait until
there are at least two or three concrete cases, unless reversal would be very
expensive later.
|
|
Introducing
patterns the team cannot maintain
|
Can the
team operate, debug, and evolve this pattern without external heroes?
|
Choose a
simpler pattern now and define the conditions that would justify moving to
the more advanced one later.
|
|
Treating
architecture as a personal project
|
Who else
needs to understand, support, and change this design?
|
Bring the
team into the decision and make ownership, trade-offs, and support
expectations explicit.
|
|
Optimizing
for an ideal end state instead of current value
|
What value
is delayed by pursuing the ideal version now?
|
Deliver a
focused slice first, then strengthen the architecture as real feedback
appears.
|
Use the diagnostic table to distinguish over-engineering, hypothetical design, premature abstraction, and patterns beyond team capability. section 3.1























