<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Coordination on Thiago Avelino</title><link>https://avelino.run/tags/coordination/</link><description>Recent content in Coordination on Thiago Avelino</description><generator>Hugo</generator><language>en-us</language><copyright>© Avelino</copyright><lastBuildDate>Fri, 24 Apr 2026 08:34:32 -0300</lastBuildDate><atom:link href="https://avelino.run/tags/coordination/index.xml" rel="self" type="application/rss+xml"/><item><title>The bug was coordination, not code</title><link>https://avelino.run/the-bug-was-coordination-not-code/</link><pubDate>Fri, 24 Apr 2026 00:00:00 +0000</pubDate><guid>https://avelino.run/the-bug-was-coordination-not-code/</guid><description>&lt;p&gt;Three teams made reasonable decisions. The product broke. That's not an engineering failure - it's what happens when you confuse process with overlap.&lt;/p&gt;
&lt;h2 id="the-postmortem-is-honest-about-the-wrong-thing"&gt;The postmortem is honest about the wrong thing&lt;/h2&gt;
&lt;p&gt;Anthropic published a postmortem on the Claude Code regression. Worth reading in full. Three separate changes, different teams, different weeks, that together looked like model degradation:&lt;/p&gt;
&lt;p&gt;Product switched default reasoning effort from &lt;code&gt;high&lt;/code&gt; to &lt;code&gt;medium&lt;/code&gt; - latency was killing the UX. Engineering shipped a caching optimization to reduce cost on session resume. Another team added a system prompt line to control verbosity on the new model.&lt;/p&gt;</description></item></channel></rss>