<?xml version="1.0" encoding="UTF-8"?>
    <rss version="2.0">
      <channel>
        <title>UseDesktop Research</title>
        <link>https://blog.usedesktop.com</link>
        <description>Research notes on verifiable RL environments, workflow trajectory data, verifier audits, and evals for computer-use agents.</description>
        <language>en-us</language>
        
        <item>
          <title>Verifier-backed workflow data for CUA</title>
          <link>https://blog.usedesktop.com/posts/verifier-backed-workflow-data-for-cua</link>
          <guid>https://blog.usedesktop.com/posts/verifier-backed-workflow-data-for-cua</guid>
          <description>CUA workflow data needs verifier audits, pass@k distributions, failure traces, and contamination controls to be useful.</description>
          <pubDate>Fri, 05 Jun 2026 00:00:00 GMT</pubDate>
        </item>
        <item>
          <title>Why computer-use agents need RL environments</title>
          <link>https://blog.usedesktop.com/posts/why-computer-use-agents-need-rl-environments</link>
          <guid>https://blog.usedesktop.com/posts/why-computer-use-agents-need-rl-environments</guid>
          <description>For computer-use agents, the bottleneck is not only RL algorithms. It is the supply of verifiable workflow environments.</description>
          <pubDate>Fri, 05 Jun 2026 00:00:00 GMT</pubDate>
        </item>
        <item>
          <title>What counts as good computer-use data?</title>
          <link>https://blog.usedesktop.com/posts/what-counts-as-good-cua-data</link>
          <guid>https://blog.usedesktop.com/posts/what-counts-as-good-cua-data</guid>
          <description>Good CUA data is workflow data with verifiers, eval splits, failure traces, and evidence that it improves computer-use agents.</description>
          <pubDate>Tue, 02 Jun 2026 00:00:00 GMT</pubDate>
        </item>
      </channel>
    </rss>