<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Vibe-Coding on danangell.com/blog</title>
    <link>https://danangell.com/blog/tags/vibe-coding/</link>
    <description>danangell.com/blog (Vibe-Coding)</description>
    <generator>Hugo -- gohugo.io</generator>
    <language>en-us</language>
    <lastBuildDate>Wed, 07 Oct 2026 23:48:00 -0700</lastBuildDate>
    
    <atom:link href="https://danangell.com/blog/tags/vibe-coding/index.xml" rel="self" type="application/rss+xml" />
    
    
    <item>
      <title>Claude gave the BrainFuck Ray Tracer a 9000x Speedup</title>
      <link>https://danangell.com/blog/posts/brainfuck-ray-tracer-optimization/</link>
      <pubDate>Wed, 07 Oct 2026 23:48:00 -0700</pubDate>
      <guid>https://danangell.com/blog/posts/brainfuck-ray-tracer-optimization/</guid>
      <description>&lt;p&gt;A few weeks ago someone published a blog post about a &lt;a href=&#34;https://epestr.com/blog/writing-a-ray-tracer-in-brainfuck/&#34;&gt;brainfuck ray tracer&lt;/a&gt;. The file is 23MB of brainfuck instructions and runs around 1 pixel per minute (when rendering the sky, which is the fastest part of the image). That means it takes &lt;em&gt;months&lt;/em&gt; to render the complete image.&lt;/p&gt;
&lt;p&gt;I wanted to see how my own &lt;a href=&#34;https://github.com/danthedaniel/BF-JIT&#34;&gt;brainfuck JIT&lt;/a&gt; handled the problem. Unsurprisingly, it produces about 1 pixel per minute. I had been gifted &amp;ldquo;unlimited&amp;rdquo; Claude tokens for an evening for an unrelated hackathon. After I finished the main project I pointed Claude Opus 5.5 at the compiler and told it to optimize &lt;code&gt;ray.bf&lt;/code&gt; - &amp;ldquo;no cheating!&amp;rdquo;. After 1 hour and 45 minutes Claude had the image rendering so much faster that a full image would only take &lt;em&gt;minutes&lt;/em&gt;! &lt;a href=&#34;https://github.com/danthedaniel/BF-JIT/tree/optimizer&#34;&gt;Here&lt;/a&gt; is the branch with Claude&amp;rsquo;s changes.&lt;/p&gt;
&lt;p&gt;I post this not to brag about work I didn&amp;rsquo;t do. But you may find yourself in the position to loop Claude over an easily evaluated problem. Give it a shot. Given the net ~1,500 extra lines of code I think the 9000x speedup is justified. If we use these tools appropriately the future of software is bright.&lt;/p&gt;
</description>
    </item>
    
  </channel>
</rss>
