<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Agentdesk on The Farah Stack</title><link>https://thefarahstack.com/tags/agentdesk/</link><description>Recent content in Agentdesk on The Farah Stack</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Mon, 05 Oct 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://thefarahstack.com/tags/agentdesk/index.xml" rel="self" type="application/rss+xml"/><item><title>Where Does Turn 12 Go? Load Balancing Agent Sessions</title><link>https://thefarahstack.com/architecture/agentdesk-load-balancing/</link><pubDate>Mon, 05 Oct 2026 00:00:00 +0000</pubDate><guid>https://thefarahstack.com/architecture/agentdesk-load-balancing/</guid><description>&lt;p&gt;&lt;img src="https://thefarahstack.com/images/posts/agentdesk-load-balancing.png" alt="Terminal card: routing turn 12 of a session with round-robin lands on a cold orchestrator that reloads turns 1-11"&gt;&lt;/p&gt;&#10;&lt;p&gt;I keep coming back to turn 12 of an agent session.&lt;/p&gt;&#10;&lt;p&gt;By turn 12 a session is carrying a lot: the whole conversation so far, whatever tools it loaded,&#10;maybe a streaming connection that&amp;rsquo;s still open to the customer&amp;rsquo;s chat window. And the most&#10;&amp;ldquo;scalable&amp;rdquo; load balancer setup there is, plain round-robin, hands that turn to whichever&#10;orchestrator instance happens to be next. Usually that&amp;rsquo;s a cold one, and it has to reload all of&#10;that context before it can say a single word back.&lt;/p&gt;</description></item><item><title>"Does It Scale?" Is Three Questions Pretending to Be One</title><link>https://thefarahstack.com/architecture/agentdesk-what-scalable-means/</link><pubDate>Fri, 02 Oct 2026 00:00:00 +0000</pubDate><guid>https://thefarahstack.com/architecture/agentdesk-what-scalable-means/</guid><description>&lt;p&gt;&lt;img src="https://thefarahstack.com/images/posts/agentdesk-what-scalable-means.png" alt="Terminal card: agentdesk at 100,000 sessions. Latency 8s (was 2s), throughput fine, availability: 2 pods restarted at 3am"&gt;&lt;/p&gt;&#10;&lt;p&gt;Someone asks &amp;ldquo;does it scale?&amp;rdquo; in a design review and everyone nods, and I&amp;rsquo;ve stopped trusting the&#10;nod. Usually each person in the room is picturing a different thing, and the system was only ever&#10;tested for one of them.&lt;/p&gt;&#10;&lt;p&gt;This is stop 1 in the Scalable Backends series. Every post threads through&#10;&lt;a href="https://thefarahstack.com/architecture/agentdesk-overview/"&gt;AgentDesk&lt;/a&gt;, a fictional platform where customer teams deploy&#10;AI agents that chat with their customers, call tools and run long sessions. It&amp;rsquo;s about to have a&#10;very good, very bad week.&lt;/p&gt;</description></item></channel></rss>