Datadog has undertaken a substantial journey in cost-aware engineering by optimizing their infrastructure, saving $17 million, and developing Cloud Cost Management for customers. The company initially focused on manual code optimization of Go functions to reduce CPU usage in high-scale services. This manual process laid the groundwork for BitsEvolve, an internal agentic system for self-optimizing code that employs evolutionary algorithms to automate performance improvements. The system uses real-world observability data to identify optimization opportunities and runs continuous evaluation loops to refine code, ultimately leading to significant cost reductions. Despite the success of manual optimizations, scaling these efforts across the organization required transitioning to an automated system that integrates observability data, AI agents, and human expertise. This approach has not only improved specific functions by up to 90% but also aims to evolve into a self-optimizing system for broader performance-critical parts of the codebase. The project underscores the importance of combining manual expertise with automated tools to achieve sustainable performance enhancements at scale.