Strangler fig in practice: modernising legacy systems without the big bang
Big-bang rewrites fail for predictable reasons. A field guide to extracting domains safely, with dual-run reconciliation patterns that satisfy auditors.
Tomás Ferreira
Principal Engineer, ByteDots
The graveyard of enterprise IT is full of big-bang rewrite tombstones: two years, eight figures, cancelled. The strangler-fig pattern remains the only approach we recommend for live revenue systems, and it works when executed with discipline.
Pick domains, not modules
Extract along business seams — quoting, billing, claims — never along technical layers. Each extracted domain needs clear ownership and its own data store.
The anti-corruption layer is the project
A translation boundary between old and new worlds protects each side from the other’s assumptions. Budget generously for it; skimping here poisons everything downstream.
Dual-run reconciliation builds trust
Run old and new implementations in parallel, diff the outputs continuously, and only cut over after parity holds above your agreed threshold. Auditors and regulators love the paper trail; executives love sleeping at night.
Retire ruthlessly
The pattern only pays off when legacy paths actually get switched off. Track decommissioning like a first-class deliverable, or you will pay twice forever.
More like this
Keep reading
The FinOps playbook we use to cut cloud bills by 30%+
Rightsizing is table stakes. Here is the five-lens framework our engineers apply to find the real savings hiding in AWS, Azure and GCP bills.
Amara Osei
Jul 14, 2026
Shipping LLM features to production: guardrails that actually work
Evaluation harnesses, prompt-injection defences and cost circuit-breakers — the checklist we run before any AI feature sees real users.
Priya Raghavan
Jun 30, 2026
SOC 2 without slowing down: controls your engineers won’t hate
Compliance projects die when they fight engineering culture. How to map SOC 2 controls onto tooling you already run, and automate the evidence.
Hannah Weiss
Apr 22, 2026
Working on something similar? This is literally what we do for a living.
Talk to us