<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
 <channel>
 <title>A True Dev</title>
 <link>https://www.atruedev.com/</link>
 <description>Field notes on agentic development - context, orchestration, verification, and the economics of delegation.</description>
 <language>en</language>
 <atom:link href="https://www.atruedev.com/rss.xml" rel="self" type="application/rss+xml" />
 <lastBuildDate>Wed, 18 Nov 2026 07:04:21 GMT</lastBuildDate>
<item><title>EXP-001 - Porting 151 posts off a Next.js blog with an agent</title><link>https://www.atruedev.com/lab/exp-001-legacy-migration/</link><guid isPermaLink="true">https://www.atruedev.com/lab/exp-001-legacy-migration/</guid><pubDate>Mon, 16 Nov 2026 08:00:00 GMT</pubDate><description>Can an agent move a large legacy content site to a new stack without losing content, URLs, or its own credibility? Run log, failures included.</description><category>migration</category><category>agents</category><category>astro</category></item>
<item><title>The verification gap</title><link>https://www.atruedev.com/writing/the-verification-gap/</link><guid isPermaLink="true">https://www.atruedev.com/writing/the-verification-gap/</guid><pubDate>Thu, 12 Nov 2026 08:00:00 GMT</pubDate><description>Writing code got cheap. Knowing whether it is correct did not. That asymmetry is now the whole job.</description><category>verification</category><category>review</category><category>agents</category></item>
<item><title>EXP-002 - Does a written plan file actually improve success?</title><link>https://www.atruedev.com/lab/exp-002-plan-file/</link><guid isPermaLink="true">https://www.atruedev.com/lab/exp-002-plan-file/</guid><pubDate>Tue, 10 Nov 2026 08:00:00 GMT</pubDate><description>Queued experiment: A/B the same refactor tasks with and without a plan artefact committed to the repo, measuring success rate and review cost.</description><category>planning</category><category>context</category><category>evals</category></item>
<item><title>Plan as artefact</title><link>https://www.atruedev.com/patterns/plan-as-artifact/</link><guid isPermaLink="true">https://www.atruedev.com/patterns/plan-as-artifact/</guid><pubDate>Sun, 08 Nov 2026 08:00:00 GMT</pubDate><description>The agent writes the plan to a file in the repository, before writing any code, and the human gate happens there.</description><category>planning</category><category>review</category></item>
<item><title>Context is a budget, not a prompt</title><link>https://www.atruedev.com/writing/context-is-a-budget/</link><guid isPermaLink="true">https://www.atruedev.com/writing/context-is-a-budget/</guid><pubDate>Thu, 05 Nov 2026 08:00:00 GMT</pubDate><description>Every token in the window costs money, latency, and attention. Treating it as free is why agents get worse the longer they work.</description><category>context</category><category>engineering</category><category>agents</category></item>
<item><title>Context budget</title><link>https://www.atruedev.com/patterns/context-budget/</link><guid isPermaLink="true">https://www.atruedev.com/patterns/context-budget/</guid><pubDate>Thu, 05 Nov 2026 08:00:00 GMT</pubDate><description>Allocate the window deliberately across layers - instructions, project facts, working set, history - instead of filling it until it overflows.</description><category>context</category><category>performance</category></item>
<item><title>Tests the agent cannot author</title><link>https://www.atruedev.com/patterns/tests-agent-cannot-author/</link><guid isPermaLink="true">https://www.atruedev.com/patterns/tests-agent-cannot-author/</guid><pubDate>Tue, 03 Nov 2026 08:00:00 GMT</pubDate><description>Write the acceptance test yourself, run it before the code exists, and never let the agent weaken it to go green.</description><category>verification</category><category>testing</category></item>
<item><title>Separate author and reviewer</title><link>https://www.atruedev.com/patterns/separate-author-and-reviewer/</link><guid isPermaLink="true">https://www.atruedev.com/patterns/separate-author-and-reviewer/</guid><pubDate>Fri, 30 Oct 2026 08:00:00 GMT</pubDate><description>A second agent with a fresh context reviews the diff against the task, not against the author&apos;s reasoning.</description><category>review</category><category>orchestration</category></item>
<item><title>Blast radius</title><link>https://www.atruedev.com/patterns/blast-radius/</link><guid isPermaLink="true">https://www.atruedev.com/patterns/blast-radius/</guid><pubDate>Mon, 26 Oct 2026 08:00:00 GMT</pubDate><description>Constrain what a run can touch, so that a wrong answer is cheap to roll back rather than expensive to discover.</description><category>safety</category><category>permissions</category></item>
<item><title>The harness matters more than the model</title><link>https://www.atruedev.com/writing/the-harness-matters-more/</link><guid isPermaLink="true">https://www.atruedev.com/writing/the-harness-matters-more/</guid><pubDate>Sat, 24 Oct 2026 08:00:00 GMT</pubDate><description>Same model, different harness, different result. The wrapper is where the compounding happens - and it is the part you can actually control.</description><category>harness</category><category>tooling</category><category>agents</category></item>
<item><title>Checkpoint and resume</title><link>https://www.atruedev.com/patterns/checkpoint-and-resume/</link><guid isPermaLink="true">https://www.atruedev.com/patterns/checkpoint-and-resume/</guid><pubDate>Thu, 22 Oct 2026 08:00:00 GMT</pubDate><description>Write progress to a file as you go, so a dead session is an inconvenience instead of a restart.</description><category>orchestration</category><category>reliability</category></item>
 </channel>
</rss>