<?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>Agent Evaluation on Thomas Talks AI: Claude Architecture, AI Agent Security, Claude Code &amp; Production Engineering</title>
    <link>https://thomasdevos.com/tags/agent-evaluation/</link>
    <description>Recent content in Agent Evaluation on Thomas Talks AI: Claude Architecture, AI Agent Security, Claude Code &amp; Production Engineering</description>
    <generator>Hugo -- 0.145.0</generator>
    <language>en-us</language>
    <copyright>Thomas De Vos ©</copyright>
    <lastBuildDate>Sun, 25 Oct 2026 13:30:00 +0000</lastBuildDate>
    <atom:link href="https://thomasdevos.com/tags/agent-evaluation/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Your agent got more accurate by answering fewer questions</title>
      <link>https://thomasdevos.com/posts/agent-abstention-risk-coverage/</link>
      <pubDate>Sun, 25 Oct 2026 13:30:00 +0000</pubDate>
      <guid>https://thomasdevos.com/posts/agent-abstention-risk-coverage/</guid>
      <description>Measure agent abstention with selective risk, coverage and escalation volume. A tested fixture shows why answered-only accuracy can hide lost service.</description>
    </item>
  </channel>
</rss>
