<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Evaluation on Programmer.ie: Modern AI programming</title>
    <link>http://programmer.ie/tags/evaluation/</link>
    <description>Recent content in Evaluation on Programmer.ie: Modern AI programming</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <lastBuildDate>Sat, 08 Aug 2026 17:31:00 +0100</lastBuildDate>
    <atom:link href="http://programmer.ie/tags/evaluation/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Agents From First Principles 09: AI Agent Says It Worked When It Didn’t? Verify the Result Outside the LLM</title>
      <link>http://programmer.ie/post/agents-from-first-principles-09/</link>
      <pubDate>Sat, 08 Aug 2026 17:31:00 +0100</pubDate>
      <guid>http://programmer.ie/post/agents-from-first-principles-09/</guid>
      <description>&lt;p&gt;An AI agent says:&lt;/p&gt;&#xA;&lt;blockquote&gt;&#xA;&lt;p&gt;Done. The task is complete.&lt;/p&gt;&lt;/blockquote&gt;&#xA;&lt;p&gt;That sentence is almost worthless.&lt;/p&gt;&#xA;&lt;p&gt;The agent may have:&lt;/p&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;edited the wrong file,&lt;/li&gt;&#xA;&lt;li&gt;changed the right file incorrectly,&lt;/li&gt;&#xA;&lt;li&gt;skipped part of the request,&lt;/li&gt;&#xA;&lt;li&gt;broken another subsystem,&lt;/li&gt;&#xA;&lt;li&gt;failed to save its work,&lt;/li&gt;&#xA;&lt;li&gt;misread a tool result,&lt;/li&gt;&#xA;&lt;li&gt;passed a stale test,&lt;/li&gt;&#xA;&lt;li&gt;inspected the wrong environment,&lt;/li&gt;&#xA;&lt;li&gt;or simply decided that its own answer looked convincing.&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;p&gt;The central problem is simple:&lt;/p&gt;&#xA;&lt;blockquote&gt;&#xA;&lt;p&gt;&lt;strong&gt;The system that produced the answer should not be the only system deciding whether the answer is correct.&lt;/strong&gt;&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
