When do you choose centralized vs decentralized AI architecture?

FILTER BY CATEGORY

ANSWER MODE

WRITTEN ANSWER

←→

### Signal to interviewer

I can set architecture boundaries that balance platform consistency with team execution speed.

### Clarify

I would clarify governance requirements, domain variance across teams, and operational risk from inconsistency.

### Approach

Apply a federated boundary model: centralize shared AI platform primitives, decentralize domain-specific orchestration and product flows.

### Metrics & instrumentation

Primary metric: team delivery throughput under policy compliance. Secondary metrics: platform reuse rate, local experiment velocity, and integration defect frequency. Guardrails: duplicated infra spend and governance drift.

### Tradeoffs

Centralization improves control and reuse but may slow teams. Decentralization increases autonomy but can create incompatible stacks and quality variance.

### Risks & mitigations

Risk: central team bottlenecks; mitigate with self-service tooling. Risk: architecture drift; mitigate with interface standards. Risk: weak ownership lines; mitigate with clear platform contracts.

### Example

A global assistant platform centralizes policy enforcement and observability while country teams control language-specific ranking and UX tuning.

### 90-second version

Use centralized architecture for shared controls and decentralized architecture for domain speed. A federated model typically delivers the best long-term balance.

FOLLOW-UPS

Clarification

Which capabilities must be centrally governed versus team-owned?
How much domain variance exists across product lines?

Depth

What API contracts prevent drift in decentralized extensions?
How would you measure when centralization is becoming a bottleneck?

Back to Interview Prep