<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0">
  <channel>
    <title>James Ross: writing</title>
    <link>https://jamesross.ai/writing/</link>
    <description>Essays on running AI agents in production, each built on something I measured in my own workspace, and a digest of the public work as it ships. New pieces also go out on Substack, where you can get them by email.</description>
    <language>en-au</language>
    <lastBuildDate>Mon, 16 Nov 2026 00:00:00 +1000</lastBuildDate>
    <item>
      <title>What an accepted task costs</title>
      <link>https://jamesross.ai/writing/what-an-accepted-task-costs</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/what-an-accepted-task-costs</guid>
      <pubDate>Mon, 16 Nov 2026 00:00:00 +1000</pubDate>
      <description>A month of agent build tasks priced by the task a reviewer accepted, and where a 3.6 times gap between two models came from.</description>
    </item>
    <item>
      <title>Digest #4: three teardowns, a new Action and six experiments</title>
      <link>https://jamesross.ai/writing/digest-4</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/digest-4</guid>
      <pubDate>Mon, 16 Nov 2026 00:00:00 +1000</pubDate>
      <description>Three teardowns, a new Action that runs the workspace check in CI, and six experiments published with their null results.</description>
    </item>
    <item>
      <title>The understanding was a by-product of the doing</title>
      <link>https://jamesross.ai/writing/the-understanding-was-a-by-product-of-the-doing</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/the-understanding-was-a-by-product-of-the-doing</guid>
      <pubDate>Sun, 18 Oct 2026 00:00:00 +1000</pubDate>
      <description>When AI does the work, the people checking it stop understanding it, and the research says a training course won’t fix that.</description>
    </item>
    <item>
      <title>You can't tell whether your AI setup got better</title>
      <link>https://jamesross.ai/writing/you-cant-tell-whether-your-ai-setup-got-better</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/you-cant-tell-whether-your-ai-setup-got-better</guid>
      <pubDate>Sun, 27 Sep 2026 00:00:00 +1000</pubDate>
      <description>Here's a cheap way to test an AI change without a research team.</description>
    </item>
    <item>
      <title>Digest #3: a teardown of GPT-RAG and two corrections</title>
      <link>https://jamesross.ai/writing/digest-3</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/digest-3</guid>
      <pubDate>Sun, 27 Sep 2026 00:00:00 +1000</pubDate>
      <description>New public work since Digest #2. Short notes here, the detail at the links.</description>
    </item>
    <item>
      <title>Stop asking the AI to be careful. Make carefulness a config file.</title>
      <link>https://jamesross.ai/writing/stop-asking-the-ai-to-be-careful</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/stop-asking-the-ai-to-be-careful</guid>
      <pubDate>Sun, 06 Sep 2026 00:00:00 +1000</pubDate>
      <description>The most reliable control in my agent workspace is a script that can't be persuaded, because it never reads the argument.</description>
    </item>
    <item>
      <title>Digest #2: three teardowns and a learn track</title>
      <link>https://jamesross.ai/writing/digest-2</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/digest-2</guid>
      <pubDate>Sun, 06 Sep 2026 00:00:00 +1000</pubDate>
      <description>Herdr, LifeOS and DeepSeek Harness read against the reference architecture, plus a free seven-module learn track.</description>
    </item>
    <item>
      <title>Your agent bill is mostly context, not output</title>
      <link>https://jamesross.ai/writing/your-agent-bill-is-mostly-context</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/your-agent-bill-is-mostly-context</guid>
      <pubDate>Sun, 30 Aug 2026 00:00:00 +1000</pubDate>
      <description>The same file read carried 147x more re-reading depending on when it loaded. That makes it an architecture decision.</description>
    </item>
    <item>
      <title>Your agents are about to start working unattended</title>
      <link>https://jamesross.ai/writing/your-agents-are-about-to-start-working-unattended</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/your-agents-are-about-to-start-working-unattended</guid>
      <pubDate>Sun, 30 Aug 2026 00:00:00 +1000</pubDate>
      <description>Nothing is watching them. Four questions that need mechanical answers before the volume arrives.</description>
    </item>
    <item>
      <title>Digest #1: three patterns, a linter, and a map</title>
      <link>https://jamesross.ai/writing/digest-1</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/digest-1</guid>
      <pubDate>Fri, 28 Aug 2026 00:00:00 +1000</pubDate>
      <description>A fortnight of shipped work on agent workspaces, and why each piece matters.</description>
    </item>
    <item>
      <title>The self-auditing workspace</title>
      <link>https://jamesross.ai/writing/the-self-auditing-workspace</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/the-self-auditing-workspace</guid>
      <pubDate>Fri, 29 May 2026 00:00:00 +1000</pubDate>
      <description>An agent workspace runs while you sleep, and rots while you sleep. So mine audits itself about once a week. The hard part wasn't building the audit. It was the three ways a self-audit quietly lies to you.</description>
    </item>
    <item>
      <title>Copilot is not the workspace</title>
      <link>https://jamesross.ai/writing/copilot-is-not-the-workspace</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/copilot-is-not-the-workspace</guid>
      <pubDate>Wed, 27 May 2026 00:00:00 +1000</pubDate>
      <description>How to put a stateless enterprise LLM assistant inside a workspace that gives it the persistence, provenance, and validation it doesn't have on its own.</description>
    </item>
    <item>
      <title>Build first, ask second</title>
      <link>https://jamesross.ai/writing/build-first-ask-second</link>
      <guid isPermaLink="true">https://jamesross.ai/writing/build-first-ask-second</guid>
      <pubDate>Sun, 24 May 2026 00:00:00 +1000</pubDate>
      <description>Turning a scheduled agent from a question-asker into a sandboxed teammate that ships work for review.</description>
    </item>
  </channel>
</rss>
