Modernizing legacy systems without a rewrite

The full rewrite is the most expensive decision in enterprise software and the most frequently chosen. It is attractive because the existing system is genuinely painful and the new one is, for now, imaginary. The failure mode is well documented: two systems to maintain, a migration that outlives its sponsors, and a feature freeze that hands the market to competitors.

Strangle, do not replace

The strangler-fig approach routes traffic through a facade in front of the legacy system, then moves capabilities behind that facade one at a time. Each move is independently valuable, independently reversible, and shippable inside a quarter. The legacy system shrinks until what remains is small enough to retire without ceremony.

The discipline it demands is choosing the first slice well. Pick something with a clear boundary, real business value, and low blast radius — rarely the most painful part of the system, which is usually the most entangled.

Build the integration core first

Most legacy estates are not one system but several, joined by nightly batch jobs and undocumented assumptions. Before extracting anything, establish a layer that exposes the data those systems own through a stable, documented interface. Everything afterwards — new services, analytics, AI — depends on this layer existing.

• Expose legacy data through explicit contracts rather than direct database reads
• Capture change events so downstream systems stop polling on a schedule
• Put the facade in front of consumers early, before any extraction begins
• Keep the legacy system authoritative until a slice is fully proven in production

Where AI actually fits

AI does not modernise a legacy system. It makes the data trapped inside one accessible — which is often the outcome the modernisation was chasing anyway.

A twenty-year-old system with decades of operational history is a genuine asset. Retrieval over its documentation and records, classification of its unstructured fields, and natural-language access to reporting all deliver value without touching the core. These are frequently the fastest wins available, and they can ship while the structural work continues underneath.

Sequence for reversibility

Every step should be one you can undo on a bad afternoon. Dual-write before you cut over. Shadow-run the new path against the old and compare outputs before trusting it. Keep the rollback switch until the new path has survived a full business cycle — a month-end close, a quarter, whatever period stresses your domain.

Modernisation done this way is undramatic. There is no launch date and no press release. There is a system that works throughout, costs less every quarter, and can finally accommodate the things the business has been asking for.