TechPrufer
Get started
Datadog
Datadog logoDatadog · Feature trail

Agentic Code Optimization

1 post · Sep 2025

What this trail covers

Datadog is developing BitsEvolve, an internal agentic system for self-optimizing code. This system leverages evolutionary algorithms to mutate code, evaluate variants against benchmarks (fitness function), and iterate. It employs an island model for parallel evolution and sharing of top performers. This builds upon manual code optimization efforts in critical hot-path functions, aiming to scale deep optimization work beyond specialized expertise.

All Datadog trails
  1. From hand-tuned Go to self-optimizing code: Building BitsEvolve | DatadogThis post details the evolution of Datadog's approach to code optimization, moving from manual, hand-tuned Go code improvements in critical hot-path functions to the development of BitsEvolve, an internal agentic system for self-optimizing code. It highlights specific manual optimizations like loop unrolling in `NormalizeTag` and aggressive fast-path optimization in `NormalizeTagArbTagValue`, which yielded significant performance gains (25% and >90% speedups respectively) and cost savings. The post then explains how these manual wins inspired the design of BitsEvolve, an evolutionary algorithm-based system that uses an island model for parallel evolution and sharing of top performers, integrating with observability and benchmarking infrastructure to scale deep optimization work.