Evolve a monolith owned by multiple autonomous engineering teams

• Identify whether the pain is deploy coupling, code ownership, runtime scaling, reliability, or team coordination. Map business domains and change dependencies before choosing process boundaries.

• Measure build and deploy time, change failure, cross-team file edits, database contention, and domain load. Runtime decomposition is not justified by team count alone.

• Teams need clear ownership and safer independent change, while preserving transactional correctness and current availability. Every extracted boundary needs a named operating team.

• Build a modular monolith with domain packages, enforced dependency direction, stable internal APIs, ownership, test seams, and schema access rules. Extract only domains with distinct scaling, failure, or delivery needs through a strangler.

• Route cross-domain commands through explicit interfaces and publish domain events through an outbox. Prohibit new direct table access across owners and inventory legacy access for removal.

• Keep local transactions where invariants are naturally shared. Before extraction, define saga, reservation, or compensation semantics rather than accidentally distributing the transaction.

• Use deployment rings, feature flags, shadow paths, and per-domain observability. Extraction adds timeouts, retries, and partial failure, so prove the benefit exceeds new operational risk.

• Make authorization and tenant context explicit at domain boundaries. A new network hop is not automatically a stronger security boundary.

• Track dependency violations, domain change lead time, deploy failure, cross-team coordination, and incident ownership. Measure extraction outcomes against the original pain.

• Modularity is cheaper than a service fleet and often captures most organizational benefit. Price new on-call, environments, networking, and data duplication before extraction.

• A modular monolith retains shared-failure and deploy coupling but preserves simple transactions. Services buy independent scale and release at the cost of distributed operations; use them selectively.

• Establish ownership and dependency checks, untangle schemas, improve deployment, then extract one evidence-backed domain. Stop if the measurable outcome does not improve.

• Senior candidates diagnose the bottleneck instead of accepting microservices as the goal. They align code and team boundaries, preserve invariants, and sequence reversible changes.

Interviewer evaluation rubric