JetBrains experimented with Google DeepMind's AlphaEvolve to optimize the B-tree implementation at the core of IntelliJ-based IDE indexing. AlphaEvolve generated and ranked algorithm variants using a synthetic benchmark loop; sessions with 50+ iterations consistently achieved 15–20% synthetic score improvements. Two of five generated candidates were validated in full IDE integration tests using Kotlin Spring Petclinic on IntelliJ IDEA 2026.2 nightly builds. One candidate reduced end-to-end indexing time from 17.4 seconds to 16.6 seconds (~4.6% improvement). The key takeaway is that autonomous search expanded the optimization search space in a mature, heavily hand-tuned codebase, while engineers retained ownership of benchmarks, code review, and release decisions. Next steps include validating improvements against the internal 'Mega Index' user-experience KPI.
