<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>The SI Arena</title><description>Long-form notes on building systems that stay up — distributed systems, data infrastructure, and the engineering decisions behind them, including the ones that turned out to be wrong.</description><link>https://thesiarena.com/</link><language>en-us</language><generator>Astro</generator><item><title>Designed agent memory stopped paying off, and the benchmarks missed it</title><link>https://thesiarena.com/post/designed-agent-memory-stopped-paying-off/</link><guid isPermaLink="true">https://thesiarena.com/post/designed-agent-memory-stopped-paying-off/</guid><description>Four papers published in October 2026 found that designed agent memory stopped beating a plain context window with current models, and that the standard benchmarks could not have told you.</description><pubDate>Sat, 24 Oct 2026 00:00:00 GMT</pubDate><category>engineering</category><category>agent-memory</category><category>evaluation</category><category>agent-benchmarks</category><category>llm-inference</category><author>Yasser</author></item></channel></rss>