The microservices invoice

Microservices were the right answer at Netflix and the wrong default everywhere else. That sentence has been written a thousand times; invoices still arrive because architecture slides travel faster than headcount models. The pitch — small teams, small services, independent deploys — is seductive to VPs who conflate repository count with autonomy.

What the pitch underprices is everything networks touch: partial failures, observability, contract versioning, local development, integration tests, on-call rotations, and the human coordination tax when no one owns the end-to-end user journey. A modular monolith fails in one process with one stack trace. A mesh fails in whispers across forty logs and a trace ID someone forgot to propagate.

SOA already taught this lesson with XML and ESB hubs. Microservices were SOA with better conference swag and worse boundaries when sold to orgs without platform teams. The ESB case studies — single throat to choke, change-control bottlenecks dressed as decoupling — are microservices prequels with WSDL costumes.

Kubernetes and Docker did not cause the fad, but they lowered the activation energy. Packaging became easy; architecture did not become free. Teams stood up clusters before they could name service ownership. Premature decomposition followed: eighty engineers, forty repos, velocity flat for two years, and a later "rationalization" project that was a monolith reunion tour with better CI.

Big-bang cutovers — flag-day migrations from monolith to mesh — combine the worst of rewrite culture with the worst of distributed ops. HeartFlow-scale healthtech and fintech pattern cases share the same postmortem shape: parity unproven, rollback compromised, observability "phase two." Regulators and customers do not care that your diagram is cloud-native.

Microservices did not fail as an idea. They failed as cargo cult. The durable move for most products is a well-tested monolith with clear module boundaries, extracting services when metrics — not slides — prove a bottleneck. That story does not keynot, which is how you know it is probably true.

The invoice line items are subtle: SRE hires, APM bills, incident trauma, slower feature shipping, and the eventual rewrite-back-to-monolith project dressed as "platform consolidation." None of that appears on the first deck that says "we need to scale like Netflix." It appears years later, when the CFO asks why infra headcount tripled while NPS flatlined.