The Observability Bill is Coming Due

AI coding agents like Cursor and GitHub Copilot are generating instrumentation code without regard for logging standards or telemetry quality, turbochargeing the long-standing garbage-in problem in observability. The result is ballooning costs (Datadog charges by volume regardless of data usefulness) and degrading signal-to-noise ratios. Sawmills, founded by veterans from Splunk, New Relic, and CloudBees, is positioning itself as an Agentic Telemetry Management Platform that sits between code and observability backends, continuously optimizing what gets sent, eliminating noise, and shifting instrumentation quality control left into the CI stage. The system keeps humans in the loop for sign-offs while an agent monitors the pipeline continuously — something a quarterly audit can never achieve given how fast AI-generated code changes the codebase.