# Study Tests Whether Language Servers Save Coding Agents Tokens and Finds They Usually Do Not

A five-arm ablation on Python and TypeScript repositories, run with three Claude models, measures tokens-to-success for semantic retrieval against plain text search.

- Published: 2026-08-18T06:24:23.025Z
- Canonical: https://polylog.news/ai/2026-08-18/study-tests-whether-language-servers-save-coding-agents-toke
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.CL](https://arxiv.org/abs/2608.13568)

Coding agents spend most of their context budget on retrieval, and the field has settled on an assumption that has almost never been measured. Semantic retrieval through the Language Server Protocol returns typed, precise results and should…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-08-18/study-tests-whether-language-servers-save-coding-agents-toke (subscription information: https://polylog.news/pricing).