<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Palanax — Writing</title>
    <link>https://www.palanax.com/writing</link>
    <atom:link href="https://www.palanax.com/writing/rss.xml" rel="self" type="application/rss+xml" />
    <description>Teardowns, cost breakdowns, and practical notes on doing AI well — including what broke.</description>
    <language>en-gb</language>
    <lastBuildDate>Thu, 11 Jun 2026 00:00:00 GMT</lastBuildDate>
    <item>
      <title>Cloud AI APIs vs. self-hosting: the 12-month cost math, done properly</title>
      <link>https://www.palanax.com/writing/cloud-vs-self-hosting-12-month-cost-math</link>
      <guid isPermaLink="true">https://www.palanax.com/writing/cloud-vs-self-hosting-12-month-cost-math</guid>
      <pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate>
      <category>Engineering teardowns</category>
      <description>&quot;Self-hosting is cheaper&quot; and &quot;the cloud is cheaper&quot; are both wrong as stated. Here&apos;s the actual break-even analysis, with the variables that decide it for your situation.</description>
    </item>
    <item>
      <title>We built a private document assistant for a law firm in three weeks. Here&apos;s what broke.</title>
      <link>https://www.palanax.com/writing/private-document-assistant-law-firm-what-broke</link>
      <guid isPermaLink="true">https://www.palanax.com/writing/private-document-assistant-law-firm-what-broke</guid>
      <pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate>
      <category>Build in public</category>
      <description>The honest version. What we scoped, what we got wrong, the retrieval problem that ate two days, and what we&apos;d do differently next time.</description>
    </item>
    <item>
      <title>What it actually costs to run an open model on your own hardware</title>
      <link>https://www.palanax.com/writing/cost-to-run-open-model-on-your-hardware</link>
      <guid isPermaLink="true">https://www.palanax.com/writing/cost-to-run-open-model-on-your-hardware</guid>
      <pubDate>Thu, 14 May 2026 00:00:00 GMT</pubDate>
      <category>Engineering teardowns</category>
      <description>We put Llama and Qwen on three hardware setups — a £600 mini-PC, a single used GPU, and a rented cloud instance — and measured throughput, latency, and cost per million tokens. Here&apos;s the real math, including where self-hosting stops making sense.</description>
    </item>
    <item>
      <title>The three AI projects every professional-services firm should do first (and the two they shouldn&apos;t)</title>
      <link>https://www.palanax.com/writing/three-ai-projects-professional-services-should-do-first</link>
      <guid isPermaLink="true">https://www.palanax.com/writing/three-ai-projects-professional-services-should-do-first</guid>
      <pubDate>Thu, 02 Apr 2026 00:00:00 GMT</pubDate>
      <category>Practical adoption</category>
      <description>We&apos;ve had the same conversation with enough law and accounting firms to know where the money is — and where it&apos;s quietly wasted. A prioritized list, with the reasoning.</description>
    </item>
    <item>
      <title>Your team is already pasting client data into ChatGPT. Here&apos;s what to do this week.</title>
      <link>https://www.palanax.com/writing/shadow-ai-what-to-do-this-week</link>
      <guid isPermaLink="true">https://www.palanax.com/writing/shadow-ai-what-to-do-this-week</guid>
      <pubDate>Thu, 05 Mar 2026 00:00:00 GMT</pubDate>
      <category>Doing it safely</category>
      <description>Shadow AI isn&apos;t a future risk; it&apos;s happening in your office right now. A calm, practical playbook — the five-step version you can start today, before you spend a rupee or a pound on tooling.</description>
    </item>
  </channel>
</rss>