<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Rainkernel insights</title>
    <link>https://rainkernel.com/insights</link>
    <atom:link href="https://rainkernel.com/rss.xml" rel="self" type="application/rss+xml" />
    <description>Sourced analysis on how AI agents fail in production and the controls that catch each.</description>
    <language>en</language>
    <lastBuildDate>Sun, 04 Oct 2026 12:17:37 GMT</lastBuildDate>
    <item>
      <title>Five ways AI agents failed in production in 2026 — and the control that would have caught each</title>
      <link>https://rainkernel.com/insights/five-ways-agents-failed-in-production-2026</link>
      <guid isPermaLink="true">https://rainkernel.com/insights/five-ways-agents-failed-in-production-2026</guid>
      <pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate>
      <description>The incidents of 2026 were not model failures. They were missing controls, and almost none of them got a CVE. Here is the catalogue, mapped to the OWASP Top 10 for Agentic Applications and to the test that finds each one before go-live.</description>
      <category>security</category>
      <category>agents</category>
      <category>OWASP</category>
    </item>
    <item>
      <title>The $50,000 hour: why agent cost needs a governor, not an alert</title>
      <link>https://rainkernel.com/insights/the-50000-dollar-hour</link>
      <guid isPermaLink="true">https://rainkernel.com/insights/the-50000-dollar-hour</guid>
      <pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate>
      <description>Every budget control sold today stops at the API key. The money is spent inside the agent loop, one retry, one escalation and one spawned sub-agent at a time — and the alert arrives a day later.</description>
      <category>cost</category>
      <category>agents</category>
      <category>FinOps</category>
    </item>
  </channel>
</rss>
