<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Ai-Agents on Ed Barros</title><link>https://barrosef.com/tags/ai-agents/</link><description>Recent content in Ai-Agents on Ed Barros</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Wed, 16 Sep 2026 17:00:00 -0400</lastBuildDate><atom:link href="https://barrosef.com/tags/ai-agents/index.xml" rel="self" type="application/rss+xml"/><item><title>When is an AI agent's work done?</title><link>https://barrosef.com/writing/when-is-an-agents-work-done/</link><pubDate>Wed, 16 Sep 2026 17:00:00 -0400</pubDate><guid>https://barrosef.com/writing/when-is-an-agents-work-done/</guid><description>&lt;p>An agent opens a pull request in twenty minutes. A person takes an hour to
review it properly. Put five agents on five demands and the bottleneck is no
longer writing the code — it is the reviewer&amp;rsquo;s desk. Parallelism that stops
there only moves the queue.&lt;/p>
&lt;p>&lt;a href="https://barrosef.com/about/">DOP&lt;/a> is a platform I am building where a developer and
an agent run a demand end to end. Two of its architecture decisions answer the
question in the title, and the answer is the same in both: &lt;strong>no green, no
PR&lt;/strong> — and green has to be earned honestly.&lt;/p></description></item></channel></rss>