<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Matthew Firth: Articles</title>
    <link>https://matthew-firth.com/articles/</link>
    <atom:link href="https://matthew-firth.com/feed.xml" rel="self" type="application/rss+xml"/>
    <description>Field notes on cloud platforms, AI governance, and building products solo.</description>
    <language>en-us</language>
    <lastBuildDate>Sat, 14 Nov 2026 12:00:00 -0500</lastBuildDate>
    <item>
      <title>ISO 42001 certification is a delta on ISO 27001</title>
      <link>https://matthew-firth.com/articles/iso-42001-certification</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/iso-42001-certification</guid>
      <pubDate>Sat, 14 Nov 2026 12:00:00 -0500</pubDate>
      <description>ISO 42001 certifies an AI management system. If you already run ISO 27001, most of it carries over. Here&#39;s what&#39;s new, when it pays off, and the cost.</description>
    </item>
    <item>
      <title>An AI acceptable use policy is only as real as its receipts</title>
      <link>https://matthew-firth.com/articles/ai-acceptable-use-policy</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/ai-acceptable-use-policy</guid>
      <pubDate>Tue, 10 Nov 2026 12:00:00 -0500</pubDate>
      <description>Most AI acceptable use policy templates are lists of rules. Give each rule a gate that enforces it and a receipt that proves it ran, or call it guidance.</description>
    </item>
    <item>
      <title>SOC 2 Type 1 vs Type 2: it&#39;s a calendar decision</title>
      <link>https://matthew-firth.com/articles/soc-2-type-1-vs-type-2</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/soc-2-type-1-vs-type-2</guid>
      <pubDate>Sun, 08 Nov 2026 12:00:00 -0500</pubDate>
      <description>SOC 2 Type 1 vs Type 2 is a calendar decision, not a rigor one. Work out when a Type 2 can exist, then decide if a Type 1 earns its fee. The window math, by date.</description>
    </item>
    <item>
      <title>ISO 27001 certification cost: price the three-year cycle</title>
      <link>https://matthew-firth.com/articles/iso-27001-cost-three-year-view</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/iso-27001-cost-three-year-view</guid>
      <pubDate>Fri, 06 Nov 2026 12:00:00 -0500</pubDate>
      <description>ISO 27001 runs on a three-year cycle, yet cost guides price the audit. Budget the cycle in three buckets, with a worksheet built from published vendor ranges.</description>
    </item>
    <item>
      <title>ISO 27001 vs SOC 2: they answer different questions</title>
      <link>https://matthew-firth.com/articles/iso-27001-vs-soc-2</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/iso-27001-vs-soc-2</guid>
      <pubDate>Wed, 04 Nov 2026 12:00:00 -0500</pubDate>
      <description>ISO 27001 certifies a security management system. SOC 2 Type II attests that controls ran over months. Pick by the question your buyers ask, not by reputation.</description>
    </item>
    <item>
      <title>Let the agent commit. Keep the push for a person.</title>
      <link>https://matthew-firth.com/articles/ai-agent-commit-push-permissions</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/ai-agent-commit-push-permissions</guid>
      <pubDate>Sat, 24 Oct 2026 12:00:00 -0400</pubDate>
      <description>How I let an AI coding agent commit and branch freely while a person keeps the push to main: a refusal in the tool, a fail-closed check, and a guarded command that failed safely.</description>
    </item>
    <item>
      <title>Which model wrote that commit? A git commit trailer is the cheapest audit control for AI code</title>
      <link>https://matthew-firth.com/articles/ai-commit-trailers-audit-evidence</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/ai-commit-trailers-audit-evidence</guid>
      <pubDate>Sat, 24 Oct 2026 12:00:00 -0400</pubDate>
      <description>A commit trailer records which AI model wrote a change, and one git command makes it checkable. The cheapest audit evidence for AI-assisted code, in place before an auditor asks.</description>
    </item>
    <item>
      <title>The deprecation notice came six months early. My tooling broke anyway.</title>
      <link>https://matthew-firth.com/articles/model-name-is-a-dependency</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/model-name-is-a-dependency</guid>
      <pubDate>Sat, 24 Oct 2026 12:00:00 -0400</pubDate>
      <description>OpenAI gave six months of notice before retiring gpt-image-1. Three of my four image tools still named it. Why the inventory, not the notice, is the control that matters.</description>
    </item>
    <item>
      <title>The shadow AI risk nobody is auditing: your own commits</title>
      <link>https://matthew-firth.com/articles/shadow-ai-code-audit-trail</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/shadow-ai-code-audit-trail</guid>
      <pubDate>Sat, 03 Oct 2026 12:00:00 -0400</pubDate>
      <description>Shadow AI is not only an unsanctioned chatbot. AI-generated code that skips the audit trail your compliance program assumes is the sharper risk. Three low-cost controls and the numbers behind them.</description>
    </item>
    <item>
      <title>Most "AI matchmaking" is a dating app with a chatbot bolted on</title>
      <link>https://matthew-firth.com/articles/ai-matchmaking-vs-dating-apps</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/ai-matchmaking-vs-dating-apps</guid>
      <pubDate>Thu, 24 Sep 2026 12:00:00 -0400</pubDate>
      <description>If an app still rewards time in app over meetings arranged, calling it AI matchmaking is marketing, not a different mechanism. A three-question test for telling the two apart.</description>
    </item>
    <item>
      <title>Why most AI memory benchmarks measure the wrong thing</title>
      <link>https://matthew-firth.com/articles/ai-memory-recall-test</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/ai-memory-recall-test</guid>
      <pubDate>Thu, 10 Sep 2026 12:00:00 -0400</pubDate>
      <description>Storage size and retrieval latency are easy to benchmark. Whether the right fact comes back when you actually need it is the number that matters, and almost nobody tests for it.</description>
    </item>
    <item>
      <title>How we cut our cloud bill in half without cutting capability</title>
      <link>https://matthew-firth.com/articles/cutting-cloud-spend-in-half</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/cutting-cloud-spend-in-half</guid>
      <pubDate>Sun, 12 Jul 2026 12:00:00 -0400</pubDate>
      <description>No heroics and no move to bare metal. A methodical audit of what we were paying for, the architectural calls that halved the bill, and the habits that keep it from growing back.</description>
    </item>
    <item>
      <title>Building NexSpark solo: what AI actually changes about being a founder</title>
      <link>https://matthew-firth.com/articles/building-nexspark-solo</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/building-nexspark-solo</guid>
      <pubDate>Sun, 28 Jun 2026 12:00:00 -0400</pubDate>
      <description>Two products, one matching engine, no employees. What held up, what didn't, and where AI earns its keep. Lessons from building NexSpark and NexSpark Friends solo.</description>
    </item>
    <item>
      <title>What 14TB of IoT telemetry taught me about boring technology</title>
      <link>https://matthew-firth.com/articles/14tb-of-telemetry-on-boring-technology</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/14tb-of-telemetry-on-boring-technology</guid>
      <pubDate>Fri, 05 Jun 2026 12:00:00 -0400</pubDate>
      <description>Why TCP, Azure Service Bus, and SQL Server are the reason a platform holding 14TB of industrial IoT telemetry works, and when I would choose something less boring.</description>
    </item>
    <item>
      <title>The strangler fig pattern in practice: nine years on a monolith 100+ Fortune 500 clients used daily</title>
      <link>https://matthew-firth.com/articles/escaping-the-monolith-without-a-rewrite</link>
      <guid isPermaLink="true">https://matthew-firth.com/articles/escaping-the-monolith-without-a-rewrite</guid>
      <pubDate>Thu, 14 May 2026 12:00:00 -0400</pubDate>
      <description>What the strangler fig pattern is, and what it looks like over nine years on a telecom expense platform 100+ Fortune 500 clients used daily: first seams, the database problem, and how to tell it is working.</description>
    </item>
  </channel>
</rss>
