<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: RAHUL ADLAKHA</title>
    <description>The latest articles on DEV Community by RAHUL ADLAKHA (@adlakharahul).</description>
    <link>https://dev.to/adlakharahul</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4252514%2F00948cfe-c372-472d-ac00-33d678a72282.jpg</url>
      <title>DEV Community: RAHUL ADLAKHA</title>
      <link>https://dev.to/adlakharahul</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/adlakharahul"/>
    <language>en</language>
    <item>
      <title>Your Agent Refunded Someone $40. Your Logs Have No Idea Why.</title>
      <dc:creator>RAHUL ADLAKHA</dc:creator>
      <pubDate>Wed, 18 Nov 2026 19:36:43 +0000</pubDate>
      <link>https://dev.to/adlakharahul/your-agent-refunded-someone-40-your-logs-have-no-idea-why-l34</link>
      <guid>https://dev.to/adlakharahul/your-agent-refunded-someone-40-your-logs-have-no-idea-why-l34</guid>
      <description>&lt;p&gt;Your support agent just issued a &lt;strong&gt;$40 refund nobody asked for&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;You check the logs. The refund tool says &lt;em&gt;"refund sent."&lt;/em&gt; The model provider shows 30 prompts that minute from random users. The guardrail logged 200 cheerful &lt;code&gt;allow&lt;/code&gt;s. Every log is correct, and none of them can tell you &lt;strong&gt;what the agent read right before it decided to refund.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;That's the core problem with agents: they choose their path at runtime, so logs that only record &lt;em&gt;outcomes&lt;/em&gt; can't explain &lt;em&gt;decisions&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;In the full post I walk through:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;🌳 &lt;strong&gt;One trace ID + a tree of spans:&lt;/strong&gt; turning three useless logs into one story you can read top to bottom&lt;/li&gt;
&lt;li&gt;🏷️ &lt;strong&gt;Why every guardrail verdict needs its policy version:&lt;/strong&gt; the difference between an explanation and a shrug&lt;/li&gt;
&lt;li&gt;🔁 &lt;strong&gt;Replay as a release gate:&lt;/strong&gt; a new guardrail replayed over 50,000 real actions caught 13 bad refunds… and would have wrongly blocked 30 legit ones&lt;/li&gt;
&lt;li&gt;🔒 &lt;strong&gt;The bill:&lt;/strong&gt; privacy, redaction and trace volume&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;👉 &lt;strong&gt;Read it here:&lt;/strong&gt; &lt;a href="https://iamrahuladlakha.github.io/blog/agent-observability-traces-replay/" rel="noopener noreferrer"&gt;Agent Observability: Traces, Replay and the Refund Nobody Asked For&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Curious: is anyone actually replaying production traffic before shipping a prompt or guardrail change? Or is it still on the roadmap next to "write better docs"? 😄&lt;/p&gt;

</description>
      <category>ai</category>
      <category>agents</category>
      <category>llm</category>
      <category>monitoring</category>
    </item>
  </channel>
</rss>
