<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Uncategorised on Tahir Hashmi</title><link>https://tahirhashmi.com/categories/uncategorised/</link><description>Recent content in Uncategorised on Tahir Hashmi</description><generator>Hugo</generator><language>en-GB</language><managingEditor>mail@tahirhashmi.com (Tahir Hashmi)</managingEditor><webMaster>mail@tahirhashmi.com (Tahir Hashmi)</webMaster><lastBuildDate>Tue, 23 Jun 2026 22:56:44 +1000</lastBuildDate><atom:link href="https://tahirhashmi.com/categories/uncategorised/index.xml" rel="self" type="application/rss+xml"/><item><title>Rebuilding Levee with Claude</title><link>https://tahirhashmi.com/posts/rebuilding-levee-with-claude-opus/</link><pubDate>Sun, 08 Mar 2026 15:48:01 +0700</pubDate><author>mail@tahirhashmi.com (Tahir Hashmi)</author><guid>https://tahirhashmi.com/posts/rebuilding-levee-with-claude-opus/</guid><description>&lt;p&gt;I had been building &lt;a href="https://github.com/codemartial/levee"

 target="_blank" rel="noopener"
&gt;levee&lt;/a&gt;, a self-configuring circuit breaker and rate limiter, on and off for almost a year. Then I over-designed it to failure. Part of the blame was mine, for trusting faulty benchmarks &amp;ndash; implemented wrongly by, yes, Claude. It was only when I started questioning the results that Claude owned up to the flaws in its own benchmark logic. After several rounds of re-architecting them for a realistic simulation, the verdict was a gut-punch: Levee was the worst performing circuit breaker of the lot, by a huge margin.&lt;/p&gt;</description></item></channel></rss>