A difficult codebase is not yet a migration plan
A team wants to rewrite an application because changes feel slow. That frustration is real, but a new repository does not explain how customers, data, integrations, and operational exceptions will move. Before choosing a rebuild, identify what must continue working throughout the transition.
Rebuild when the replacement is sufficiently bounded, the behaviour is understood, and migration can be rehearsed. Modernize incrementally when business activity cannot pause and useful boundaries allow old and new parts to coexist. Neither approach is inherently more disciplined. Both can fail when the real system is defined only as its visible screens.
What a rebuild buys and what it forgets
A rebuild can remove structural constraints that are genuinely expensive to work around. It offers room to simplify a confused domain model or replace unsupported foundations. Those benefits should be named precisely rather than reduced to a preference for a newer framework.
The hidden risk is forgotten behaviour. Export formats, permission exceptions, reconciliation jobs, and support procedures may be undocumented but essential. A clean implementation that omits them can be less complete than the system it replaces.
Keep a behaviour inventory and distinguish requirements that must survive from habits that should change. Avoid promising a feature-for-feature copy until the team understands what the existing system actually does. Reproducing accidental complexity is not modernization.
What incremental replacement requires
Incremental work needs a boundary at which responsibility can move safely. That might be a workflow, a module, or a public interface. The old and new implementation must agree about data ownership and how requests are routed.
AWS describes the strangler fig pattern as gradual replacement and highlights routing, coupling, and data consistency concerns. Its examples use microservices, but the useful decision principle is controlled replacement. A modernization project does not have to become a microservices project.
Coexistence is not free. Temporary adapters, synchronization, and duplicate operational paths need owners and removal conditions. If each migrated part leaves another permanent bridge behind, the transition can increase complexity instead of reducing it.
Treat data and operations as first-class scope
Identify the authoritative writer for each record during the transition. Decide how changes made after a migration snapshot reach the destination, how disagreements are detected, and what rollback means once new transactions exist.
Also inventory links, scheduled jobs, integrations, and support access. A redesigned interface does not compensate for broken external URLs or an overnight process that no longer runs. Test the operating day, not just the login and homepage.
Clodron's web platform engineering should be evaluated against this transition work as well as the new application. The questions in what a partner should explain before signing help expose who owns migration, rollback, and post-release support. A proposal that only describes the destination is incomplete.
Find a reversible slice this week
Select a meaningful workflow and map its dependencies, data changes, and acceptance evidence. Ask whether it can move independently without creating conflicting sources of truth.
For a rebuild candidate, rehearse migration with a safe copy of representative data and demonstrate how the old system could remain available if the cutover fails. For incremental work, demonstrate routing and a clear way to retire the replaced component.
Choose the path that makes progress observable and failure containable. If no safe boundary can be found, investigate the domain and dependencies before estimating a complete replacement. The goal is a system the business can keep using while its future becomes easier to change.
