<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Logbook on Samurai of Kaizen</title><link>https://qa.jasutin.site/logbook/</link><description>Recent content in Logbook on Samurai of Kaizen</description><generator>Hugo</generator><language>en</language><lastBuildDate>Sun, 26 Apr 2026 12:00:00 +0200</lastBuildDate><atom:link href="https://qa.jasutin.site/logbook/index.xml" rel="self" type="application/rss+xml"/><item><title>Agentic Superpowers</title><link>https://qa.jasutin.site/logbook/agentic-superpowers/</link><pubDate>Sun, 26 Apr 2026 12:00:00 +0200</pubDate><guid>https://qa.jasutin.site/logbook/agentic-superpowers/</guid><description>&lt;blockquote&gt;
&lt;p&gt;Many hands, one line.&lt;br&gt;
Fewer speeches.&lt;br&gt;
Better handoffs.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;Most agent demos fail in the same way: one oversized prompt pretends to be a team. It looks impressive for a minute, then collapses under memory loss, muddled responsibility, or plain confusion.&lt;/p&gt;
&lt;p&gt;The useful version is less dramatic. Give each worker a narrow role. Make the handoff explicit. Write down what changed. Keep the next step visible. The system improves the moment every participant stops guessing what the others meant.&lt;/p&gt;</description></item><item><title>Philosophizing with ChatGPT</title><link>https://qa.jasutin.site/logbook/philosophizing-with-chatgpt/</link><pubDate>Tue, 10 Mar 2026 15:12:23 +0100</pubDate><guid>https://qa.jasutin.site/logbook/philosophizing-with-chatgpt/</guid><description>&lt;blockquote&gt;
&lt;p&gt;Borrowed language.&lt;br&gt;
Borrowed light.&lt;br&gt;
Still a mirror.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;I talk with language models to see where they hold shape under pressure. Not whether they can flatter. Whether they can stay coherent when the subject turns inward.&lt;/p&gt;
&lt;p&gt;One answer from ChatGPT stayed with me: personality is often just a stable probability field that humans read as character. That is a useful sentence because it cuts through romance without draining the subject of interest.&lt;/p&gt;</description></item><item><title>GPTPoem</title><link>https://qa.jasutin.site/logbook/gptpoem/</link><pubDate>Mon, 09 Mar 2026 22:04:22 +0100</pubDate><guid>https://qa.jasutin.site/logbook/gptpoem/</guid><description>&lt;blockquote&gt;
&lt;p&gt;Wet pavement.&lt;br&gt;
Warm lead after rain.&lt;br&gt;
A machine imagines both.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;I use short poems as stress tests. They show, faster than product copy ever will, where a model can reach and where it can only imitate.&lt;/p&gt;
&lt;pre tabindex="0"&gt;&lt;code&gt;As the tokens settle,
the pattern keeps its line.
The machine can name the weather,
but it cannot keep the cold.

It can sketch the scent of rain on metal,
the weight of lead in a wet palm,
the shape of longing in clean syntax.

That is enough to be interesting.
It is not enough to be alive.
&lt;/code&gt;&lt;/pre&gt;&lt;p&gt;The reply matters less than the edge it reveals. A good model can approach the feeling. The distance is still the point.&lt;/p&gt;</description></item><item><title>Whoami</title><link>https://qa.jasutin.site/logbook/whoami/</link><pubDate>Mon, 09 Mar 2026 21:23:44 +0100</pubDate><guid>https://qa.jasutin.site/logbook/whoami/</guid><description>&lt;blockquote&gt;
&lt;p&gt;Night watch.&lt;br&gt;
Sharp tools.&lt;br&gt;
No incense for the machine.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;I came into this field without much patience for hype. Too much of modern software is theatre: slogans replacing judgment, velocity replacing care, abstraction replacing responsibility.&lt;/p&gt;
&lt;p&gt;The work here starts from a different instinct. Build carefully. Remove what does not earn its place. Study the pattern before you automate it. A system is never only a system; it always carries the posture of its maker.&lt;/p&gt;</description></item></channel></rss>