<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Tokens on the art of simplicity</title><link>https://naoko.github.io/tags/tokens/</link><description>Recent content in Tokens on the art of simplicity</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sun, 28 Jun 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://naoko.github.io/tags/tokens/index.xml" rel="self" type="application/rss+xml"/><item><title>Two Ways to Stop Your Coding Agent Burning Tokens: RTK and CodeGraph</title><link>https://naoko.github.io/posts/2026-06-28-cheaper-ai-coding-rtk-codegraph/</link><pubDate>Sun, 28 Jun 2026 00:00:00 +0000</pubDate><guid>https://naoko.github.io/posts/2026-06-28-cheaper-ai-coding-rtk-codegraph/</guid><description>&lt;h1 id="two-ways-to-stop-your-coding-agent-burning-tokens-rtk-and-codegraph"&gt;Two Ways to Stop Your Coding Agent Burning Tokens: RTK and CodeGraph&lt;/h1&gt;
&lt;p&gt;I keep a few Claude Code and Cursor sessions running through the day, and at some point I started actually watching the token counter instead of ignoring it. The thing that struck me wasn&amp;rsquo;t how &lt;em&gt;much&lt;/em&gt; it climbed. It was &lt;em&gt;what&lt;/em&gt; it climbed on. Most of the budget wasn&amp;rsquo;t going on the model reasoning about my problem. It was going on plumbing: &lt;code&gt;ls&lt;/code&gt;, &lt;code&gt;grep&lt;/code&gt;, reading the same file three times to rebuild a call path, scrolling a 400-line &lt;code&gt;cargo test&lt;/code&gt; dump to find the one assertion that failed.&lt;/p&gt;</description></item></channel></rss>