John Demic developed OpenAI Streaming API Token Usage Tracking for the mckinsey/agents-at-scale-ark repository, focusing on enhancing observability and cost accounting for streaming API calls. He implemented token usage collection by enabling the IncludeUsage option in streaming requests and capturing usage data from the final streaming chunk. This approach addressed the persistent issue of zero token counts in OTEL traces, resulting in more accurate cost tracking and improved debugging capabilities. Working primarily with Go, John applied his expertise in API integration and backend development to deliver higher data quality and more reliable telemetry for streaming workloads, with minimal disruption to existing flows.