<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Agentic Harness on Bruce on AI Engineering</title><link>http://www.heyuan110.com/tags/agentic-harness/</link><description>Recent content in Agentic Harness on Bruce on AI Engineering</description><generator>Hugo</generator><language>en</language><lastBuildDate>Sun, 05 Jul 2026 10:00:00 +0800</lastBuildDate><atom:link href="http://www.heyuan110.com/tags/agentic-harness/index.xml" rel="self" type="application/rss+xml"/><item><title>Loop Engineering: Building the Cage Your AI Agent Runs In</title><link>http://www.heyuan110.com/posts/ai/2026-07-05-loop-engineering/</link><pubDate>Sun, 05 Jul 2026 10:00:00 +0800</pubDate><guid>http://www.heyuan110.com/posts/ai/2026-07-05-loop-engineering/</guid><description>&lt;p&gt;&lt;img src="http://www.heyuan110.com/posts/ai/2026-07-05-loop-engineering/cover.webp"
 alt="Loop engineering: the four-pillar cage an AI agent runs inside in 2026"
 
 loading="lazy"
 decoding="async"
 fetchpriority="low"
 width="1200"
 height="630"
/&gt;
&lt;/p&gt;
&lt;p&gt;Here is the reframe that should change how you build agents in 2026: everyone is optimizing the animal, and the durable engineering is in the cage. The model is the animal — powerful, fast-improving, and increasingly a commodity you rent by the token. &lt;strong&gt;Loop engineering&lt;/strong&gt; is the cage: the control loop that decides when to trim the agent&amp;rsquo;s memory, when to slam the brakes, when to reject its own homework, and which tools it is allowed to touch. Swap Opus for GPT-5.x for Gemini and your agent gets a little smarter. Build the wrong loop and any of them will happily burn a $200 bill overnight and hand you a red CI. My core claim: &lt;strong&gt;the model is water and electricity; the loop and its guardrails are the moat.&lt;/strong&gt;&lt;/p&gt;</description></item></channel></rss>