<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>Engineering on AI Pulse</title>
    <link>https://cuigh.com/tags/engineering/</link>
    <description>Recent content in Engineering on AI Pulse</description>
    <generator>Hugo</generator>
    <language>en-US</language>
    <lastBuildDate>Mon, 13 Jul 2026 09:51:14 +0800</lastBuildDate>
    <atom:link href="https://cuigh.com/tags/engineering/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Migrating a production AI agent from Claude Opus 4.8 to GPT-5.6 Sol: the three engineering costs behind the 2.2x benchmark</title>
      <link>https://cuigh.com/posts/gpt-5-6-sol-production-migration-2026/</link>
      <pubDate>Mon, 13 Jul 2026 09:51:14 +0800</pubDate>
      <guid>https://cuigh.com/posts/gpt-5-6-sol-production-migration-2026/</guid>
      <description>Ploy publicly migrated its production agent from Claude Opus 4.8 to GPT-5.6 Sol and got the visible win — 2.2x faster, 27% cheaper, 0.970 vs 0.936 visual. The three engineering costs nobody talks about are: tool-argument overreach (GPT-5.6 fills all 25 optional params by default, 52%–64% of file reads came back empty), prompt-cache design cliff (partial-prefix implicit caching was dropped, zero hit without explicit key&#43;breakpoint), and reasoning-replay mismatch (server-side item references break under multi-worker append-only input). Migrating a production agent is a stack migration, not a model swap.</description>
    </item>
  </channel>
</rss>
