Design for failure anticipates the operational risks of generative-AI services and applies established cloud and mission-critical architecture practices instead of relying on optimistic execution. section 1.7 section 1.8
A resilient enterprise integration should account for:
- Latency: commercially available generative-AI APIs are complex and slower than traditional APIs. section 1.7
- Transient failures: apply retry mechanics, including exponential backoff policies. section 1.7
- Repeated work: use caching where appropriate to reduce avoidable calls and improve operational behavior. section 1.7
- Enterprise exposure: retain security practices and extend existing architecture guidance to new AI endpoints. section 1.7 section 1.8
Enterprise API integration
An enterprise adding a generative-AI endpoint to an existing line-of-business system should preserve its current architecture practices, then add retry and exponential-backoff handling, caching, and security controls for the new dependency. section 1.7 section 1.8
