Is a monolith outdated?
No. It is a deployment shape, and many current products use it. It is a sound choice for most new products and for teams without a need to deploy parts independently.
What is a modular monolith?
A single deployable application organized into clear modules with defined interfaces and no shared database tables between modules. It keeps the simplicity of one unit and makes a later split easier.
Can a monolith scale?
Yes. You can run several copies behind a load balancer, add caching and read replicas, and move heavy jobs to background workers. Splitting out one part is an option when that is not enough.
Is a monolith cheaper to build?
Usually, at the start. There is less infrastructure, fewer pipelines and simpler testing. Costs rise later if the code is allowed to tangle, so structure matters.
How do I know if my monolith has become a problem?
Releases are slow and risky, teams wait on each other, builds take too long, or a small change causes wide failures. If those are true and structure cannot fix them, consider extracting parts.
Should a startup begin with a monolith?
Almost always. A startup is still learning what to build, and a single codebase makes change fast and cheap. Keep modules clean so you can split later if growth demands it.
What is the difference between a monolith and a legacy system?
A monolith describes structure: one deployable unit. Legacy describes age and difficulty to change. A new monolith is not legacy, and a legacy system may or may not be a monolith.
Is a monolith good for AI features?
Yes, for most products. An AI feature can start as a module inside the monolith that calls a hosted model through a thin interface. If it later needs separate scaling, such as heavy background processing, you can move it into its own worker without redesigning everything.
What is the strangler pattern?
A way to replace a system gradually. New services take over one piece of the old application at a time behind a routing layer, until what remains is small enough to leave alone or retire. It avoids a risky all-at-once rewrite.
How do I move from a monolith to microservices?
Gradually. Clean up module boundaries, put a routing layer in front, extract one well-bounded service, move traffic in steps and repeat only where the pain justifies it. Many teams stop after one or two services.