Use idempotency keys for retryable writes
Make a retried write replay the same operation instead of creating a duplicate side effect.
Reliability
Make logs structured before adding dashboards
Capture queryable fields first so dashboards are built on data, not text search.
Use IHttpClientFactory
Create configured HttpClient instances through the factory instead of scattering clients and handlers.
Do not hide exceptions inside fire-and-forget tasks
If work can fail, give it ownership, logging, cancellation, and a lifecycle.
Nullable warnings are design feedback, not noise
Treat nullable warnings as pressure to make object state and API contracts clearer.
Pass CancellationToken through real boundaries
Thread cancellation through every expensive operation instead of dropping it mid-chain.
Prefer health checks that test dependencies deliberately
Health checks should say something useful about readiness without turning every probe into production load.
Use OpenTelemetry before you need production debugging
Add traces, metrics, and logs while the system is still easy to reason about.
Why CancellationToken Matters More in .NET AI Systems
Why CancellationToken matters more in .NET AI systems, and how to pass it through LLM calls, streaming responses, embeddings, retrieval, and tool execution.