<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Yann Orieult — Writing</title><description>AI Systems Engineer — GenAI, distributed systems and production architecture. I build applied AI systems that hold up in production, and I host this site myself.</description><link>https://orieult.com/</link><language>en-us</language><atom:link href="https://orieult.com/rss.xml" rel="self" type="application/rss+xml"/><item><title>Five ways my nightly agent system failed while every run looked successful</title><link>https://orieult.com/writing/five-ways-nightly-agents-failed-while-reporting-success/</link><guid isPermaLink="true">https://orieult.com/writing/five-ways-nightly-agents-failed-while-reporting-success/</guid><description>I use agents across software, data, infrastructure, and knowledge work. Five unattended failures taught me why process success is not outcome evidence.</description><pubDate>Sat, 29 Aug 2026 00:00:00 GMT</pubDate><category>AI agents</category><category>MCP</category><category>self-hosted</category><category>reliability</category><category>observability</category></item></channel></rss>