<?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>Posts on Samantha Jayasinghe</title><link>http://samanthajayasinghe.github.io/post/</link><description>Recent content in Posts on Samantha Jayasinghe</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sun, 11 Oct 2026 00:00:00 +0000</lastBuildDate><atom:link href="http://samanthajayasinghe.github.io/post/index.xml" rel="self" type="application/rss+xml"/><item><title>Testing AI Agents: An Eval Harness That Gives You Permission to Change</title><link>http://samanthajayasinghe.github.io/post/testing-ai-agents/</link><pubDate>Sun, 11 Oct 2026 00:00:00 +0000</pubDate><guid>http://samanthajayasinghe.github.io/post/testing-ai-agents/</guid><description>Agents are stochastic, so a single passing run proves nothing. A proper test harness turns that uncertainty into evidence you can act on — including the confidence to switch models.</description></item></channel></rss>