How Netflix Made AI Coding Agents More Efficient with JetBrains IDE Intelligence - The JetBrains Blog

Netflix integrated JetBrains’ headless Java/Kotlin code-intelligence engine, based on IntelliJ IDEA’s indexing capabilities, with Claude Code and Codex to help coding agents navigate JVM projects and compiled JAR dependencies more accurately. Because the LSP server alone was not suited to repeated agent calls, Netflix created a Java-based daemon and CLI that keep project indexes warm, avoiding repeated startup and indexing costs. Across benchmarks on an internal Spring Boot service and the open-source DGS Framework, agents using the intelligence layer generally consumed less time and fewer tokens while maintaining or improving correctness, with typical efficiency gains of 20–30% and a best case near 65%, especially for stronger models operating at higher effort settings. End-to-end programming tasks showed consistent gains of roughly 10–30%, while navigation tasks reduced token use by about 15–60%, although one batched-search configuration did not show representative time savings. Netflix rejected alternatives requiring a running IDE or relying on Eclipse’s Java LSP, plans to test similar methods for other languages, and JetBrains argues that future agent tooling should expose program-level concepts such as symbols, diagnostics, tests, and runtime behavior rather than primarily relying on text-position-based LSP interactions.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.