The Cargo Cult of Microservices
Netflix uses microservices. Google uses microservices. Amazon wrote the book on microservices. Therefore you should use microservices.
This logic is the source of an enormous amount of wasted engineering time and unnecessarily complex systems.
Netflix has thousands of engineers. You have three. The problem microservices solve — scaling independent teams and services — is not a problem you have at team sizes under 50.
What Microservices Actually Cost You
Microservices introduce a distributed systems problem into every feature you ship. Service discovery. Network timeouts. Distributed tracing. Cross-service transactions. Data consistency across service boundaries. These are hard problems that require senior engineers and significant tooling to manage.
A monolith with a clean module structure has none of these problems. A function call does not fail with a network timeout.
When to Start With a Monolith
Always. Build a monolith until you can identify a specific service that:
- Has independent scaling requirements from the rest of the application
- Has a clear ownership boundary with a dedicated team
- Has a deployment frequency that is blocked by the monolith release cycle
If you cannot name the service and the specific problem it solves by being extracted, you are building microservices for aesthetic reasons.
The Modular Monolith Is the Real Best Practice
A well-structured monolith with clear module boundaries — where the user module cannot directly import from the billing module's internal functions, for example — gives you 80% of the organizational benefits of microservices with none of the distributed systems overhead.
When a module genuinely needs to become a service, the extraction is clean because the boundaries were defined from the start.
The Right Time to Extract
You have earned the microservices conversation when:
- A specific piece of functionality is deployed 10+ times per day independently
- A team of 5+ engineers owns a clear domain boundary
- Scaling that function independently would meaningfully reduce infrastructure cost
Until then, ship features in your monolith. The architecture conversation will find you when it needs to.