The hidden tax
We instrumented our internal time tracking across three years and 14 microservices. Each service averaged 3.2 weeks of pure overhead per year — auth integration, deployment scripts, on-call rotations, monitoring setup, dependency updates.
That's 3.2 weeks per service per year, paid forever, before anyone shipped a feature.
When it's worth it
Microservices are right when you have hard scaling boundaries, real team independence, or compliance reasons to isolate. They're not right when "we'll need to scale eventually" is the strongest argument.
What we recommend instead
Modular monoliths. Hard module boundaries inside a single deploy. The discipline of microservices, the cost of a single service.