{"id":630,"date":"2026-10-17T23:37:58","date_gmt":"2026-10-17T23:37:58","guid":{"rendered":"https:\/\/gilnas.synology.me\/wordpress\/630\/"},"modified":"2026-10-17T23:37:58","modified_gmt":"2026-10-17T23:37:58","slug":"ai-%ec%97%90%ec%9d%b4%ec%a0%84%ed%8a%b8-%ec%9a%b4%ec%98%81%ec%9d%98-%ec%a4%91%ec%8b%ac-%ec%9d%b4%eb%8f%99-%ed%8f%89%ea%b0%80%c2%b7%ea%b6%8c%ed%95%9c%c2%b7%eb%b3%b5%ea%b5%ac-agent-evaluation-contro","status":"publish","type":"post","link":"https:\/\/gilnas.synology.me\/wordpress\/630\/","title":{"rendered":"AI \uc5d0\uc774\uc804\ud2b8 \uc6b4\uc601\uc758 \uc911\uc2ec \uc774\ub3d9: \ud3c9\uac00\u00b7\uad8c\ud55c\u00b7\ubcf5\uad6c | Agent Evaluation, Control and Recovery"},"content":{"rendered":"
\"AI<\/figure>\n

\uc694\uc57d:<\/strong> \ucd5c\uadfc 72\uc2dc\uac04\uc758 AI\/LLM\u00b7AI Agent\u00b7\uac1c\ubc1c\uc790 \ub3c4\uad6c \ud750\ub984\uc740 \uc5d0\uc774\uc804\ud2b8\ub97c \uc2e0\ub8b0 \uac00\ub2a5\ud558\uac8c \uc6b4\uc601\ud558\ub294 \ubc29\ubc95<\/strong>\uc5d0 \ucd08\uc810\uc774 \ubaa8\uc600\ub2e4. \ud3c9\uac00\ub97c CI\uc640 \ubc30\ud3ec \uc774\ud6c4\uae4c\uc9c0 \uc5f0\uacb0\ud558\uace0, \uad8c\ud55c\u00b7\uc2b9\uc778\u00b7\uad00\uce21\u00b7\ubcf5\uad6c\ub97c \ub7f0\ud0c0\uc784 \uc124\uacc4\uc5d0 \ub123\ub294 \ubcc0\ud654\uac00 AWS, Codex, Claude Code, \uc0cc\ub4dc\ubc15\uc2a4 \ud504\ub85c\uc81d\ud2b8\uc5d0\uc11c \ubc18\ubcf5 \ud655\uc778\ub410\ub2e4.<\/p>\n

\ud575\uc2ec \ub2f5\ubcc0<\/h2>\n

\uc9c0\uae08 \uac1c\ubc1c\ud300\uc774 \uc6b0\uc120\ud560 \uc77c\uc740 \uc0c8 \uc5d0\uc774\uc804\ud2b8 \uae30\ub2a5\uc744 \uace7\ubc14\ub85c \uc5c5\ubb34 \uad8c\ud55c\uc5d0 \uc5f0\uacb0\ud558\ub294 \uac83\uc774 \uc544\ub2c8\ub77c, \ud3c9\uac00 \uae30\uc900\u00b7\ub3c4\uad6c \uad8c\ud55c\u00b7\uc2e4\ud328 \ubcf5\uad6c<\/strong>\ub97c \ud568\uaed8 \ubc30\ud3ec \ub2e8\uc704\ub85c \ub9cc\ub4dc\ub294 \uac83\uc774\ub2e4. \uc5d0\uc774\uc804\ud2b8\ub294 \uc5ec\ub7ec \ub3c4\uad6c\uc640 \uc138\uc158\uc744 \uc870\uc728\ud558\ub294 \uc2e4\ud589 \uacc4\uce35\uc774\ubbc0\ub85c \uc815\ub2f5\ub960\ubfd0 \uc544\ub2c8\ub77c \ub3c4\uad6c \uc120\ud0dd, \uad8c\ud55c \uc704\ubc18, \ube44\uc6a9, \ub4dc\ub9ac\ud504\ud2b8, \uc885\ub8cc \ucc98\ub9ac\ub97c \uce21\uc815\ud574\uc57c \ud55c\ub2e4.<\/p>\n

3\uc77c \uacb0\ub860: \uacbd\uc7c1\ub825\uc740 \ubaa8\ub378\ubcf4\ub2e4 \ud3c9\uac00 \ub8e8\ud504\uc640 \uc81c\uc5b4\uba74\uc5d0\uc11c \uac08\ub9b0\ub2e4<\/h2>\n

\uc774\ubc88 \uae30\uac04\uc5d0 \ud655\uc778\ud55c 7\uac1c \uc774\uc288\ub294 \ub124 \uac00\uc9c0 \ud750\ub984\uc73c\ub85c \ubb36\uc778\ub2e4. \uccab\uc9f8, \uc6b4\uc601 \ud488\uc9c8\uc740 \uc9c0\uc5f0\uc2dc\uac04\uacfc \uc624\ub958\uc728\ub9cc\uc73c\ub85c \uad00\ub9ac\ud560 \uc218 \uc5c6\ub2e4\ub294 \uc778\uc2dd\uc774 \uad6c\uccb4\uc801\uc778 \ud3c9\uac00 \ud30c\uc774\ud504\ub77c\uc778\uc73c\ub85c \ubc14\ub00c\uace0 \uc788\ub2e4. \ub458\uc9f8, \ucf54\ub529 \uc5d0\uc774\uc804\ud2b8\ub294 \ub300\ud654\ud615 \ub3c4\uad6c\ub97c \ub118\uc5b4 \ubcd1\ub82c \uc870\uc0ac\u00b7\ubc31\uadf8\ub77c\uc6b4\ub4dc \ub7f0\ud0c0\uc784\u00b7\uc138\uc158 \uc81c\uc5b4\ub97c \uac16\ucd98 \uc791\uc5c5 \ud658\uacbd\uc73c\ub85c \ud655\uc7a5\ud558\uace0 \uc788\ub2e4. \uc14b\uc9f8, \uc7a5\uae30 \uc2e4\ud589\uc744 \uc704\ud55c \ucee8\ud14d\uc2a4\ud2b8 \uc555\ucd95\uacfc \uc790\ub3d9 \ubd84\ub958\ub294 \ud3b8\uc758 \uae30\ub2a5\uc774\uc9c0\ub9cc \uba54\ubaa8\ub9ac\u00b7MCP \ud638\ud658\uc131\u00b7\uad00\uce21 \uc815\ubcf4\uae4c\uc9c0 \ud568\uaed8 \ub2e4\ub904\uc57c \ud55c\ub2e4. \ub137\uc9f8, \uc0cc\ub4dc\ubc15\uc2a4\uc640 \uc885\ub8cc \ucc98\ub9ac\ub294 \uc608\uc678\uac00 \uc544\ub2c8\ub77c \uc5d0\uc774\uc804\ud2b8 \uc2e0\ub8b0\uc131\uc758 \uae30\ubcf8 \uacbd\uacc4\ub2e4.<\/p>\n

1. \ud3c9\uac00\ub97c \ub9c8\uc9c0\ub9c9 QA\uac00 \uc544\ub2cc \ubc30\ud3ec \uc2dc\uc2a4\ud15c\uc73c\ub85c \ub9cc\ub4e4\uae30<\/h2>\n

\ud655\uc778\ub41c \uc0ac\uc2e4<\/h3>\n

AWS\ub294 10\uc6d4 16\uc77c AgentCore Evaluations \uae30\ubc18\uc758 9\ub2e8\uacc4 \uc5d0\uc774\uc804\ud2b8 \ud3c9\uac00 \ud504\ub808\uc784\uc6cc\ud06c<\/a>\ub97c \uacf5\uac1c\ud588\ub2e4. \ud488\uc9c8 \uc815\uc758, \uc2e4\ud5d8 \ube44\uad50, CI \ud68c\uadc0 \uac80\uc0ac, CD \ud488\uc9c8 \uac8c\uc774\ud2b8, \uce74\ub098\ub9ac, \uc801\ub300\uc801 \ud14c\uc2a4\ud2b8, \ub4dc\ub9ac\ud504\ud2b8 \uac10\uc2dc, \uc2e4\uc2dc\uac04 \uad00\uce21, \ud53c\ub4dc\ubc31 \ub8e8\ud504\ub97c \uc5f0\uc18d \uacfc\uc815\uc73c\ub85c \uc5f0\uacb0\ud55c\ub2e4. AWS\uac00 \uac15\uc870\ud55c \ud575\uc2ec\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \ud2c0\ub824\ub3c4 \uc9c0\uc5f0\uc2dc\uac04\uacfc \uc624\ub958\uc728\uc740 \uc815\uc0c1\ucc98\ub7fc \ubcf4\uc77c \uc218 \uc788\ub294 \u2018\uc870\uc6a9\ud55c \uc2e4\ud328\u2019\ub2e4.<\/p>\n

\uac19\uc740 \uae30\uac04 \uacf5\uac1c\ub41c Trust in Depth<\/a>\ub294 \ubc94\uc704 \uc124\uc815, \uc778\uc9c0, \uacbd\uacc4, \uad00\uce21, \uac80\uc99d\uc744 \uacc4\uce35\uc801\uc73c\ub85c \uc801\uc6a9\ud558\ub294 \uc6b4\uc601 \ubc29\uc2dd\uc744 \uc81c\uc548\ud588\ub2e4. \uc774\ub294 \uc5d0\uc774\uc804\ud2b8 \uad8c\ud55c\uc744 \uc804\ubd80 \ud5c8\uc6a9\ud558\uac70\ub098 \uc804\ubd80 \ub9c9\ub294 \uc774\ubd84\ubc95\ubcf4\ub2e4, \uac80\uc99d \uac00\ub2a5\ud55c \ud589\ub3d9\uc5d0\ub9cc \uc790\uc728\uc131\uc744 \ubd80\uc5ec\ud558\ub294 \uc124\uacc4\uc5d0 \uac00\uae5d\ub2e4.<\/p>\n

\uc2e4\ubb34 \uc2dc\uc0ac\uc810<\/h3>\n

\uc5d0\uc774\uc804\ud2b8 \ud488\uc9c8\uc740 \ub2e8\uc77c LLM-as-a-judge \uc810\uc218\ub85c \ub05d\ub098\uc9c0 \uc54a\ub294\ub2e4. \ub3c4\uad6c\ubcc4 \uc815\ub2f5 \uacc4\uc57d, \uc5d0\uc774\uc804\ud2b8\ubcc4 golden set, \ub2e4\uc911 \uc5d0\uc774\uc804\ud2b8 \ub77c\uc6b0\ud305\uacfc \uc885\ud569 \uacb0\uacfc\ub97c \ubd84\ub9ac\ud574 \ubd10\uc57c \ud55c\ub2e4. \uac80\uc0c9\u00b7\ucf54\ub4dc \uc218\uc815\u00b7\ubc30\ud3ec\ucc98\ub7fc \uc678\ubd80 \uc0c1\ud0dc\ub97c \ubc14\uafb8\ub294 \uc5c5\ubb34\ub294 \uc815\ud655\ub3c4, \uadfc\uac70\uc131, \uad8c\ud55c \uc900\uc218, \ube44\uc6a9, \ubcf5\uad6c \uac00\ub2a5\uc131<\/strong>\uc744 \uac19\uc740 \uc2b9\uaca9 \uae30\uc900\uc73c\ub85c \ubb36\ub294 \ud3b8\uc774 \uc548\uc804\ud558\ub2e4.<\/p>\n