As it becomes increasingly common to delegate code fixes to AI, have you ever experienced a situation where “the strategy sounds great, but not a single line of code actually changes”?
\nI experienced precisely this recently.<\/p>\n
To align with updated specifications, a large-scale refactoring affecting the entire codebase was required, so I instructed Codex to make the fixes. However, all that came back were:<\/p>\n
This state persisted, with actual files remaining completely untouched.<\/p>\n
In this article, I refer to this state\u2014where the AI explains correction strategies and cautions in detail, but fails to execute actual code changes or run commands\u2014as “Policy Artisan Mode”<\/strong> (Houshin Shokunin Mode). I will break down what was happening and how I handled it in my environment.<\/p>\n *Please note that the contents described here are based solely on the results in my own environment, and the same method is not guaranteed to reproduce universally.<\/em><\/p>\n When the state I call “Policy Artisan Mode” occurred this time, the tasks instructed to Codex were as follows:<\/p>\n This was not merely adding a function or making a minor tweak; Because doing it manually would take time, I decided to leave this task to Codex. For day-to-day development, I used the following settings:<\/p>\n The reason was simple: saving tokens.<\/p>\n Even with this configuration,<\/p>\n It was fully practical for purposes like these, and I had felt no particular issues.<\/p>\n However, for fixes like this one, Codex’s behavior was clearly different.<\/p>\n This state of “showing strategy and ideas, but not lifting a finger” is what this article calls “Policy Artisan Mode”<\/strong>.<\/p>\n The important point is that this was not due to a lack of permissions. This is purely speculation, but organizing the situation leads me to think the following:<\/p>\n Under these conditions, Codex may have concluded:<\/p>\n “There is a high probability I cannot finish this completely in this environment.” And thus gave up on actual code updates.<\/p>\n Rather than a bug, this behavior felt closer to an overly cautious self-judgment.<\/p>\n Therefore, I changed the settings as follows:<\/p>\n Once I did this, the behavior changed completely.<\/p>\n The work that had previously stalled began moving forward normally.<\/p>\n What can be said:<\/p>\n What cannot be said:<\/p>\n It is simply an observational fact that If you encounter a situation where:<\/p>\n Then trying to:<\/p>\n is a countermeasure worth trying.<\/p>\n Especially when accompanied by large-scale refactoring or heavy testing, In this case, Codex entered the so-called “Policy Artisan Mode,” In my environment, reviewing the model and resource settings seemed to get it out of that state, However, if you encounter a similar situation, As it becomes increasingly common to delegate code fixes to AI, have you ever experienced a situation where &# […]<\/p>\n","protected":false},"author":4,"featured_media":2517,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_locale":"en_US","_original_post":"https:\/\/donguri3.net\/?p=2515","footnotes":""},"categories":[1169],"tags":[540],"class_list":["post-5303","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-local-llm","tag-540","en-US"],"_links":{"self":[{"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/posts\/5303","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/comments?post=5303"}],"version-history":[{"count":1,"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/posts\/5303\/revisions"}],"predecessor-version":[{"id":5306,"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/posts\/5303\/revisions\/5306"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/media\/2517"}],"wp:attachment":[{"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/media?parent=5303"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/categories?post=5303"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/donguri3.net\/wp-json\/wp\/v2\/tags?post=5303"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}The Situation Where the Problem Occurred<\/h2>\n
\n
\nit was “work that required not just thinking, but actually getting hands dirty and proceeding through verification.”<\/p>\n
\nSince it had worked without issues for routine, small-scale fixes,
\nI had anticipated that it could handle this scale as well.<\/p>\nUsual Settings<\/h2>\n
\n
\n
The Problem: “Policy Artisan Mode”<\/h2>\n
\n
\nIn the terminal, the settings had been configured from the start to allow modifying actual files.<\/p>\nWhy It Entered Policy Artisan Mode (Speculation)<\/h2>\n
\n
\n
\n“It is safer to stop at presenting a strategy rather than touching it halfway.”<\/p>\n<\/blockquote>\nAttempted Solution: Changing Model and Resource Settings<\/h2>\n
\n
\n
What Can and Cannot Be Said From This Result<\/h2>\n
\n
\n
\n“the behavior changed drastically before and after changing the settings.”<\/p>\nHints When Encountering the Same Situation<\/h2>\n
\n
\n
\nsettings aimed at saving tokens may end up working against you.<\/p>\nConclusion: AI Also Has “Prerequisites for Working”<\/h2>\n
\nwhere it would show a strategy but leave the code unupdated.<\/p>\n
\nthough I do not know if this is a universal solution.<\/p>\n
\nreviewing your settings may provide a helpful clue.<\/p>\n","protected":false},"excerpt":{"rendered":"