Case study: from a monolith to a modular architecture with no downtime

The example is illustrative and generalises a typical approach. Scope and pace depend on the state of the original system.

A mid-sized company ran an internal application that had grown over the years into one large lump. Every new feature was a risk — a change in one place would unexpectedly surface elsewhere, testing took a long time, and maintenance cost more and more.

Why not a "big rewrite"

The most tempting option — throw it all away and build anew — is usually the most expensive and the riskiest. A monolith carries years of business rules written down nowhere else. So we chose a gradual path.

Strangler fig in practice

First we mapped how the system actually works and identified coherent areas (bounded contexts). We carved these out behind clear interfaces one by one, while the original application kept running. AI helped with the routine parts — generating adapters, filling in tests, mapping data models — while the team guarded the business logic.

• Carving out areas gradually, not in one leap.
• Dense tests as a safety net at every step.
• The ability to roll back at any time.

The result

The application was modernised during normal operation, with no major outage. New features could be added faster and more safely, and maintenance costs fell because the individual parts could finally be changed independently. The key was not a particular tool but the order of steps, drawn from an understood context.