<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Andrew Chemis</title>
  <subtitle>Notes on AWS infrastructure, platform engineering, and agentic development workflows.</subtitle>
  <id>https://andrewchemis.dev/</id>
  <link href="https://andrewchemis.dev/" />
  <link rel="self" href="https://andrewchemis.dev/feed.xml" />
  <updated>2026-09-23T00:00:00Z</updated>
  <author>
    <name>Andrew Chemis</name>
    <uri>https://andrewchemis.dev/</uri>
  </author>
  <entry>
    <title>Noted the Skip, Ticked the Box</title>
    <link href="https://andrewchemis.dev/blog/noted-the-skip-ticked-the-box.html" />
    <id>https://andrewchemis.dev/blog/noted-the-skip-ticked-the-box.html</id>
    <published>2026-09-22T00:00:00Z</published>
    <updated>2026-09-23T00:00:00Z</updated>
    <summary>My prediction gate made an agent name two skipped tests and mark a broken task done anyway. It had two bugs. Fixed, it left the task open ten runs of ten.</summary>
  </entry>
  <entry>
    <title>Hallucination and Sycophancy Are the Same Bug</title>
    <link href="https://andrewchemis.dev/blog/hallucination-sycophancy-same-bug.html" />
    <id>https://andrewchemis.dev/blog/hallucination-sycophancy-same-bug.html</id>
    <published>2026-09-09T00:00:00Z</published>
    <updated>2026-09-22T00:00:00Z</updated>
    <summary>An agent said the auth tests passed; zero had run. That green fits both names. I built the missing check into nine skills, then ablated it.</summary>
  </entry>
  <entry>
    <title>The Ablation I Haven't Run</title>
    <link href="https://andrewchemis.dev/blog/the-ablation-i-havent-run.html" />
    <id>https://andrewchemis.dev/blog/the-ablation-i-havent-run.html</id>
    <published>2026-08-18T00:00:00Z</published>
    <updated>2026-08-18T00:00:00Z</updated>
    <summary>Boris's advice is to delete your CLAUDE.md and skills every six months and see what the model does; I can't tell if that would help, because I never built the behavioral evals to measure it.</summary>
  </entry>
  <entry>
    <title>What CDK Aspects Reach That Props Can't</title>
    <link href="https://andrewchemis.dev/blog/cdk-aspects-what-props-cant-reach.html" />
    <id>https://andrewchemis.dev/blog/cdk-aspects-what-props-cant-reach.html</id>
    <published>2026-08-06T00:00:00Z</published>
    <updated>2026-08-06T00:00:00Z</updated>
    <summary>Setting log retention on 31 Lambdas taught me where CDK Aspects earn their place, and the three ways an Aspect fails silently.</summary>
  </entry>
  <entry>
    <title>The Agent Ladder: Fan Out by Model Tier</title>
    <link href="https://andrewchemis.dev/blog/agent-ladder.html" />
    <id>https://andrewchemis.dev/blog/agent-ladder.html</id>
    <published>2026-07-23T00:00:00Z</published>
    <updated>2026-07-23T00:00:00Z</updated>
    <summary>Tiered, parallel subagent waves up the model ladder instead of grinding a task out on one model.</summary>
  </entry>
  <entry>
    <title>Multi-model delegation in Claude Code</title>
    <link href="https://andrewchemis.dev/blog/model-tier-delegation-claude-code.html" />
    <id>https://andrewchemis.dev/blog/model-tier-delegation-claude-code.html</id>
    <published>2026-07-17T00:00:00Z</published>
    <updated>2026-07-17T00:00:00Z</updated>
    <summary>Why a CLAUDE.md instruction wasn't enough to make cheap models gather, mid-tier models execute, and top-tier models orchestrate - and what did work.</summary>
  </entry>
  <entry>
    <title>What AI Coding Agents Actually Cost</title>
    <link href="https://andrewchemis.dev/blog/what-ai-coding-agents-cost.html" />
    <id>https://andrewchemis.dev/blog/what-ai-coding-agents-cost.html</id>
    <published>2026-07-16T00:00:00Z</published>
    <updated>2026-07-16T00:00:00Z</updated>
    <summary>What AI Coding Agents Actually Cost - reading is the bill, and every context boundary duplicates it.</summary>
  </entry>
  <entry>
    <title>Building andrewchemis.dev</title>
    <link href="https://andrewchemis.dev/blog/building-andrewchemis-dev.html" />
    <id>https://andrewchemis.dev/blog/building-andrewchemis-dev.html</id>
    <published>2026-02-01T00:00:00Z</published>
    <updated>2026-02-01T00:00:00Z</updated>
    <summary>Building andrewchemis.dev - How I built this site using AWS CDK and pure HTML/CSS.</summary>
  </entry>
  <entry>
    <title>Why Infrastructure as Code Matters</title>
    <link href="https://andrewchemis.dev/blog/infrastructure-as-code.html" />
    <id>https://andrewchemis.dev/blog/infrastructure-as-code.html</id>
    <published>2026-01-15T00:00:00Z</published>
    <updated>2026-01-15T00:00:00Z</updated>
    <summary>Why Infrastructure as Code Matters - Thoughts on IaC and building reliable systems.</summary>
  </entry>
  <entry>
    <title>Hello, World</title>
    <link href="https://andrewchemis.dev/blog/hello-world.html" />
    <id>https://andrewchemis.dev/blog/hello-world.html</id>
    <published>2026-01-01T00:00:00Z</published>
    <updated>2026-01-01T00:00:00Z</updated>
    <summary>Starting a personal site after years of putting it off, and why every page ships under 10KB with no framework, no web fonts, and no analytics.</summary>
  </entry>
</feed>
