<?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>Flash Model on Vinh Thang's AI Tech Notes</title><link>https://vinhthang.dev/tags/flash-model/</link><description>Recent content in Flash Model on Vinh Thang's AI Tech Notes</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Tue, 22 Sep 2026 20:00:00 +0700</lastBuildDate><atom:link href="https://vinhthang.dev/tags/flash-model/index.xml" rel="self" type="application/rss+xml"/><item><title>Attention Guard Evolution: Adaptive Workflows, Lifecycle Reuse, and Flash-First Agent Governance</title><link>https://vinhthang.dev/posts/attention-guard-adaptive-workflows-and-lifecycle-evolution/</link><pubDate>Tue, 22 Sep 2026 20:00:00 +0700</pubDate><guid>https://vinhthang.dev/posts/attention-guard-adaptive-workflows-and-lifecycle-evolution/</guid><description>&lt;p&gt;In our earlier architecture deep-dives, we examined how &lt;strong&gt;Attention Dilution&lt;/strong&gt; cripples autonomous coding agents. As context windows swell past tens of thousands of tokens with terminal traces, compiler diagnostics, and speculative file edits, an agent’s attention mechanism degrades. Critical system constraints, architectural decision records (ADRs), and testing invariants drift into the periphery, leading to hallucinations, ignored instructions, and silent error suppression.&lt;/p&gt;
&lt;p&gt;To solve this in Google Antigravity, we built the &lt;strong&gt;&lt;a href="https://github.com/vinhthang/antigravity-attention-guard-plugin"&gt;Antigravity Attention Guard Plugin&lt;/a&gt;&lt;/strong&gt;—a deterministic runtime firewall that enforces subagent delegation, sandboxed tool execution, and cryptographic ledger tracking.&lt;/p&gt;</description></item></channel></rss>