<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Claude Code on Jack Monte, AI Engineer</title><link>https://jackmonte.com/tags/claude-code/</link><description>Recent content in Claude Code on Jack Monte, AI Engineer</description><image><title>Jack Monte, AI Engineer</title><url>https://jackmonte.com/images/og-image.png</url><link>https://jackmonte.com/images/og-image.png</link></image><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sat, 15 Aug 2026 18:00:00 -0700</lastBuildDate><atom:link href="https://jackmonte.com/tags/claude-code/index.xml" rel="self" type="application/rss+xml"/><item><title>My Sub-Agent Followed an Instruction I Never Wrote</title><link>https://jackmonte.com/posts/sub-agent-instruction-i-never-wrote/</link><pubDate>Sat, 15 Aug 2026 18:00:00 -0700</pubDate><guid>https://jackmonte.com/posts/sub-agent-instruction-i-never-wrote/</guid><description>Sub-agents are sold on context isolation, and that part holds up. Less obvious is that three separate inputs cross into a sub-agent&amp;#39;s context, and one of them is composed by a model on your behalf and never appears in your transcript. Here is which input won, and what delegation actually costs in tokens.</description></item></channel></rss>