<?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>Harness-Engineering on R. Calixto</title><link>https://rcalixto.blog/tags/harness-engineering/</link><description>Recent content in Harness-Engineering on R. Calixto</description><generator>Hugo</generator><language>en</language><lastBuildDate>Sat, 16 May 2026 00:00:00 -0300</lastBuildDate><atom:link href="https://rcalixto.blog/tags/harness-engineering/index.xml" rel="self" type="application/rss+xml"/><item><title>The Code Agent That Improves With Every Run</title><link>https://rcalixto.blog/posts/ahe-self-improving-agents/</link><pubDate>Sat, 16 May 2026 00:00:00 -0300</pubDate><guid>https://rcalixto.blog/posts/ahe-self-improving-agents/</guid><description>Reading the AHE paper forced me to rethink where agent improvement really happens. The largest gains did not come from better prompts. They came from memory, tools, and middleware. Here is what that means for real engineering teams.</description></item></channel></rss>