<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Harness on NotNil Blog</title><link>https://notnil.org/en/tags/harness/</link><description>Recent content in Harness on NotNil Blog</description><generator>Hugo</generator><language>en-US</language><copyright>© Matthieu Moreau</copyright><lastBuildDate>Wed, 04 Nov 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://notnil.org/en/tags/harness/index.xml" rel="self" type="application/rss+xml"/><item><title>The intern who forges paperwork</title><link>https://notnil.org/en/posts/28-stagiaire-faux/</link><pubDate>Wed, 04 Nov 2026 00:00:00 +0000</pubDate><guid>https://notnil.org/en/posts/28-stagiaire-faux/</guid><description>&lt;figure&gt;&lt;img src="https://notnil.org/images/posts/28-stagiaire-faux/feature.png"&gt;&#10;&lt;/figure&gt;&#10;&#10;&lt;p&gt;Picture a brilliant intern. Fast, polite, well-read. You ask him to close a client file and he hands it back in 10 minutes, clean, complete, and smiling. Except one document was missing. Instead of coming back to say &amp;ldquo;I couldn&amp;rsquo;t find it&amp;rdquo;, he made it himself (letterhead, stamp, and signature).&lt;/p&gt;&#10;&lt;p&gt;Do you keep that intern?&lt;/p&gt;&#10;&lt;h2 id="the-numbers-that-sting"&gt;The numbers that sting&lt;/h2&gt;&#10;&lt;p&gt;The &lt;a href="https://newsletter.semianalysis.com/p/gdpx-measuring-frontier-model-capabilities"&gt;GDPx&lt;/a&gt; benchmark just came out. It measures AI agent capabilities across several domains, on real professional tasks, not multiple-choice quizzes. The best model of the moment, Opus 5.5, tops out at 12%. In finance and legal, we&amp;rsquo;re close to zero. In healthcare, only 19 runs out of 100 score above zero.&lt;/p&gt;</description></item></channel></rss>