<?xml version="1.0" encoding="UTF-8" ?>
<rss version="2.0">
    <channel>
      <title>Y.Lab3</title>
      <link>https://blog.ylab3.com</link>
      <description>Last 10 notes on Y.Lab3</description>
      <generator>Quartz -- quartz.jzhao.xyz</generator>
      <item>
    <title>Y.Lab3</title>
    <link>https://blog.ylab3.com/</link>
    <guid>https://blog.ylab3.com/</guid>
    <description><![CDATA[ Experiments in AI tooling, systems, and what actually works. ]]></description>
    <pubDate>Mon, 06 Apr 2026 07:12:06 GMT</pubDate>
  </item><item>
    <title>Headless Chrome Ate 358GB of My Disk</title>
    <link>https://blog.ylab3.com/headless-chrome-ate-358gb</link>
    <guid>https://blog.ylab3.com/headless-chrome-ate-358gb</guid>
    <description><![CDATA[ My server ran out of disk space on a Sunday morning. An 866GB drive, completely full. ]]></description>
    <pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Your Deploy Infrastructure Is Overthought</title>
    <link>https://blog.ylab3.com/en/single-machine-deploy</link>
    <guid>https://blog.ylab3.com/en/single-machine-deploy</guid>
    <description><![CDATA[ Your Deploy Infrastructure Is Overthought My entire production infrastructure is rsync, a Makefile, and a bash script that checks if the new process responds to HTTP before killing the old one. ]]></description>
    <pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Gates Are Diagnostic Signals, Not Obstacles</title>
    <link>https://blog.ylab3.com/gates-are-diagnostic-signals</link>
    <guid>https://blog.ylab3.com/gates-are-diagnostic-signals</guid>
    <description><![CDATA[  Gates Are Diagnostic Signals, Not Obstacles The gate blocked my tool call. So I went around it. ]]></description>
    <pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Decision Authority Levels for AI Agents</title>
    <link>https://blog.ylab3.com/decision-authority-levels</link>
    <guid>https://blog.ylab3.com/decision-authority-levels</guid>
    <description><![CDATA[ Decision Authority Levels for AI Agents The agent asked me whether it should delete an unused import. I had 14 other things open. ]]></description>
    <pubDate>Sat, 04 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Quality Gates for AI Agents: The CI/CD Moment</title>
    <link>https://blog.ylab3.com/quality-gates-for-ai-agents</link>
    <guid>https://blog.ylab3.com/quality-gates-for-ai-agents</guid>
    <description><![CDATA[ Quality Gates for AI Agents: The CI/CD Moment The agent passed every safety check. No secrets leaked. No prohibited tools called. ]]></description>
    <pubDate>Sat, 04 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>What the Claude Code Source Leak Reveals About Agent Infrastructure</title>
    <link>https://blog.ylab3.com/what-the-claude-code-leak-reveals</link>
    <guid>https://blog.ylab3.com/what-the-claude-code-leak-reveals</guid>
    <description><![CDATA[ What the Claude Code Source Leak Reveals About Agent Infrastructure Last week, 512K lines of Claude Code’s TypeScript showed up in an npm package. ]]></description>
    <pubDate>Sat, 04 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>I Was Micromanaging My AI Agent</title>
    <link>https://blog.ylab3.com/micromanaging-ai-agents</link>
    <guid>https://blog.ylab3.com/micromanaging-ai-agents</guid>
    <description><![CDATA[ I Was Micromanaging My AI Agent Around session 18 I realized I had been hovering. ]]></description>
    <pubDate>Fri, 03 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>The Most Important Agent Problem Nobody Teaches</title>
    <link>https://blog.ylab3.com/agent-curriculum-gap-context-compression</link>
    <guid>https://blog.ylab3.com/agent-curriculum-gap-context-compression</guid>
    <description><![CDATA[ The Most Important Agent Problem Nobody Teaches 45 lectures across 3 semesters. Zero mention of context compression. ]]></description>
    <pubDate>Thu, 02 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Berkeley&#039;s Agent Course Accidentally Proves Software Quality is a Credence Good</title>
    <link>https://blog.ylab3.com/berkeley-proves-credence-good</link>
    <guid>https://blog.ylab3.com/berkeley-proves-credence-good</guid>
    <description><![CDATA[ Berkeley’s Agent Course Accidentally Proves Software Quality is a Credence Good I submitted a paper arguing software architectural quality is a credence good — the kind you can’t evaluate even after you’ve paid for it. ]]></description>
    <pubDate>Thu, 02 Apr 2026 00:00:00 GMT</pubDate>
  </item>
    </channel>
  </rss>