---
type: "raw_signal"
id: "improved-token-efficiency-for-longer-agent-runs-2bfa90"
slug: "improved-token-efficiency-for-longer-agent-runs-2bfa90"
title: "Improved token efficiency for longer agent runs"
description: "Cursor says it reduced users’ token costs by 7% without reducing agent quality by reworking its agent harness for longer runs."
retrieval_nugget: "Cursor says it reduced users’ token costs by 7% without reducing agent quality by reworking its agent harness for longer runs."
observed_at: "2026-09-25"
record_date: "2026-09-25"
date_kind: "observed_at"
why_it_matters: "Cursor says it A/B tested tool-context configurations against token use, cost, latency, tool-call errors, and overall agent usage. It retained common reading, search, editing, and shell tools in static context while moving other definitions to dynamic loading. The post reports a 60% reduction in static-context tool-description tokens and a 20% lower cold-cache-miss rate."
novelty: "notable"
verification_level: "source-linked"
signal_type: "release"
source_platform: "cursor.com"
topics: ["agent-harness"]
entities: []
related_patterns: []
source_url: "https://cursor.com/blog/improved-token-efficiency"
source_urls: ["https://cursor.com/blog/improved-token-efficiency"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "signal:improved-token-efficiency-for-longer-agent-runs-2bfa90"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/signals/improved-token-efficiency-for-longer-agent-runs-2bfa90/","markdown":"https://newruntime.com/signals/improved-token-efficiency-for-longer-agent-runs-2bfa90.md","json":"https://newruntime.com/signals/improved-token-efficiency-for-longer-agent-runs-2bfa90.json"}
---

# Improved token efficiency for longer agent runs

## Retrieval answer

Cursor says it reduced users’ token costs by 7% without reducing agent quality by reworking its agent harness for longer runs.

Cursor says it reduced users’ token costs by 7% without reducing agent quality by reworking its agent harness for longer runs. The company trimmed its system prompt by roughly 66%, dynamically loads less frequently needed built-in tools, improved prompt-cache placement, and reduced line-number overhead in file reads.

Cursor says it A/B tested tool-context configurations against token use, cost, latency, tool-call errors, and overall agent usage. It retained common reading, search, editing, and shell tools in static context while moving other definitions to dynamic loading. The post reports a 60% reduction in static-context tool-description tokens and a 20% lower cold-cache-miss rate.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Agent Harness](https://newruntime.com/topics/agent-harness.md): Explore the agent-harness topic hub.
- collection: [Raw Signals](https://newruntime.com/signals.json): Open every source-linked raw signal.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)