<?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>Antigravity on Vinh Thang's AI Tech Notes</title><link>https://vinhthang.dev/categories/antigravity/</link><description>Recent content in Antigravity on Vinh Thang's AI Tech Notes</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Tue, 01 Sep 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://vinhthang.dev/categories/antigravity/index.xml" rel="self" type="application/rss+xml"/><item><title>Attention Dilution: Why 1M-Token Context Windows Kill Strict Agent Behavior (And How We Fixed It)</title><link>https://vinhthang.dev/posts/attention-dilution-context-windows-agents/</link><pubDate>Tue, 01 Sep 2026 00:00:00 +0000</pubDate><guid>https://vinhthang.dev/posts/attention-dilution-context-windows-agents/</guid><description>&lt;p&gt;There is a pervasive myth in modern AI engineering: &lt;em&gt;“Just give the model a 1,000,000-token context window, paste all your enterprise governance rules in the system prompt, and let it build your system.”&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;If you have built real-world autonomous coding agents operating inside complex cloud infrastructures, multi-module monorepos, or production Kubernetes clusters, you already know the harsh truth: massive context windows don&amp;rsquo;t make agents smarter; they make them amnesiac.&lt;/p&gt;
&lt;p&gt;As an agentic conversation unfolds—accumulating terminal outputs, stack traces, file diffs, and conversational turns—a subtle failure mode emerges: &lt;strong&gt;Attention Dilution&lt;/strong&gt;.&lt;/p&gt;</description></item></channel></rss>