Feature flags, on AWS, at a tenth of the price.

LaunchDarkly’s feature set is huge; this customer used roughly a quarter of it. They had flags, percentage rollouts, gradual ramps, and audit logs. They didn’t use experimentation, segmentation by 17 dimensions, or the SDK-side analytics. The premium tier paid for everything anyway.

AppConfig with the local agent matched the runtime semantics that mattered. We built a thin SDK wrapper that translated the team’s existing flag-evaluation calls (which were already abstracted behind a custom interface) to AppConfig under the hood. Gradual rollouts mapped to AppConfig’s deployment strategies.

Shadow ran for two weeks with both systems returning evaluations. Once parity was confirmed, we switched the production reads to AppConfig and turned off LaunchDarkly. The savings funded two engineering hires in the next budget cycle.