<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>The Carbon Layer</title><description>Carbon-based humans, in a silicon world. Deep dives on how AI systems actually get built.</description><link>https://thecarbonlayer.com/</link><item><title>A meta-harness has to derive the evidence it acts on</title><link>https://thecarbonlayer.com/meta-harness-build/</link><guid isPermaLink="true">https://thecarbonlayer.com/meta-harness-build/</guid><description>An agent told my prototype it had committed a fix, the commit didn&apos;t exist, and the only thing that caught it was code that asked Git instead of asking the agent.</description><pubDate>Sun, 02 Aug 2026 00:00:00 GMT</pubDate></item><item><title>A meta-harness is coordination that lives outside the agents</title><link>https://thecarbonlayer.com/meta-harness/</link><guid isPermaLink="true">https://thecarbonlayer.com/meta-harness/</guid><description>I run four agent sessions to get one change reviewed properly and the coordination between them is still done by hand; writing down what that glue actually does produced eight components, none of which belong in a supervisor model&apos;s context.</description><pubDate>Sun, 26 Jul 2026 00:00:00 GMT</pubDate></item><item><title>A domain agent is narrow enough to measure</title><link>https://thecarbonlayer.com/agent-evaluation/</link><guid isPermaLink="true">https://thecarbonlayer.com/agent-evaluation/</guid><description>The same agent answered one question straight from the database and confidently invented the next; a 25-question eval harness is how I found out which failures matter and what to fix first.</description><pubDate>Mon, 20 Jul 2026 00:00:00 GMT</pubDate></item><item><title>The harness can improve without touching the model</title><link>https://thecarbonlayer.com/self-evolving-harness/</link><guid isPermaLink="true">https://thecarbonlayer.com/self-evolving-harness/</guid><description>An outside model found a real bug in my agent&apos;s harness, wrote a fix, and proved it helped; every safety property in that sentence had to be built by hand.</description><pubDate>Thu, 16 Jul 2026 00:00:00 GMT</pubDate></item><item><title>A working coding agent is about 2,500 lines of harness</title><link>https://thecarbonlayer.com/building-an-agent-from-scratch/</link><guid isPermaLink="true">https://thecarbonlayer.com/building-an-agent-from-scratch/</guid><description>Fifteen chapters after the empty file, the model&apos;s weights hadn&apos;t changed once; everything that reads as agentic came from code we wrote around one API call.</description><pubDate>Thu, 16 Jul 2026 00:00:00 GMT</pubDate></item><item><title>Context management decides what the model sees this turn</title><link>https://thecarbonlayer.com/context-management/</link><guid isPermaLink="true">https://thecarbonlayer.com/context-management/</guid><description>Forty turns into a real bug the agent started contradicting itself; the model never changed, the window did.</description><pubDate>Thu, 16 Jul 2026 00:00:00 GMT</pubDate></item><item><title>Agent skills are procedures the harness can load</title><link>https://thecarbonlayer.com/agent-skills/</link><guid isPermaLink="true">https://thecarbonlayer.com/agent-skills/</guid><description>Your strongest reviewer cannot sit inside every pull request; a skill captures the repeatable part of their procedure so the agent stops rediscovering the job.</description><pubDate>Thu, 16 Jul 2026 00:00:00 GMT</pubDate></item><item><title>Agent sandboxing is a blast radius decision</title><link>https://thecarbonlayer.com/agent-sandboxing/</link><guid isPermaLink="true">https://thecarbonlayer.com/agent-sandboxing/</guid><description>The word sandbox covers five very different mechanisms, from a V8 isolate to a microVM, and the right one depends on the failure you need to contain.</description><pubDate>Thu, 16 Jul 2026 00:00:00 GMT</pubDate></item><item><title>Agent memory is context assembly over time</title><link>https://thecarbonlayer.com/agent-memory-architecture/</link><guid isPermaLink="true">https://thecarbonlayer.com/agent-memory-architecture/</guid><description>The model forgets everything at the session boundary; whatever feels like memory is the harness deciding what to keep, recall, update, and forget.</description><pubDate>Thu, 16 Jul 2026 00:00:00 GMT</pubDate></item><item><title>Harness engineering is the system around the model</title><link>https://thecarbonlayer.com/harness-engineering/</link><guid isPermaLink="true">https://thecarbonlayer.com/harness-engineering/</guid><description>When an agent fails, the model takes the blame; the more useful questions are about the ten layers wrapped around it.</description><pubDate>Thu, 16 Jul 2026 00:00:00 GMT</pubDate></item></channel></rss>