![\"\"](\"https:\/\/pirago.vn\/wp-content\/uploads\/2025\/09\/image-10.png\")
<\/figure>\n\n\n\n1. Context Engineering \u306e4\u3064\u306e\u67f1<\/strong><\/h2>\n\n\n\n1.1 Write\uff08\u66f8\u304d\u51fa\u3057\uff09\u2014 context window \u306e\u5916\u306b\u60c5\u5831\u3092\u4fdd\u5b58\u3059\u308b<\/strong><\/h3>\n\n\n\n\u8ab2\u984c:<\/h4>\n\n\n\n
LLMs \u306b\u306f context window<\/em> \u306e\u4e0a\u9650\u304c\u3042\u308a\u307e\u3059\u3002 \u4f8b\u3048\u3070 GPT-4 \u306f 128k tokens\u3001Claude \u306f 200k tokens\uff08\u7d04 15,000 \u884c\u306e\u30b3\u30fc\u30c9\uff09\u307e\u3067\u3067\u3059\u3002 \u8907\u96d1\u306a\u30bf\u30b9\u30af\u3067\u306f\u91cd\u8981\u60c5\u5831\u304c\u6ea2\u308c\u3066\u3057\u307e\u3046\u3053\u3068\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n\u89e3\u6c7a\u7b56:<\/h4>\n\n\n\n
\u5916\u90e8\u30b9\u30c8\u30ec\u30fc\u30b8\u306b\u8a08\u753b\u3084\u5c65\u6b74\u3092\u4fdd\u5b58\u3057\u3001\u5fc5\u8981\u306a\u3068\u304d\u306b\u8aad\u307f\u623b\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
\u4f8b1\uff1aProject Planner Agent<\/strong> \u5916\u90e8\u30d5\u30a1\u30a4\u30eb\u306b\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u8a08\u753b\u3092\u4fdd\u5b58\u30fb\u53d6\u5f97\u3002<\/p>\n\n\n\n![\"\"](\"https:\/\/pirago.vn\/wp-content\/uploads\/2025\/09\/image-11.png\")
<\/figure>\n\n\n\n\u4f8b2\uff1aCustomer Support Agent<\/strong> \u904e\u53bb\u306e\u4f1a\u8a71\u5c65\u6b74\u3092 SQLite \u306b\u4fdd\u5b58\u3057\u3001\u30e6\u30fc\u30b6\u30fc\u3054\u3068\u306e\u6587\u8108\u3092\u547c\u3073\u51fa\u3059\u3002<\/p>\n\n\n\n![\"\"](\"https:\/\/pirago.vn\/wp-content\/uploads\/2025\/09\/image-12.png\")
<\/figure>\n\n\n\n\u52b9\u679c:<\/h4>\n\n\n\n\n- context overflow \u306e\u30a8\u30e9\u30fc\u3092 40% \u6e1b\u5c11<\/strong><\/li>\n\n\n\n
- \u8907\u96d1\u306a\u30bf\u30b9\u30af\u8a08\u753b\u306e\u7cbe\u5ea6\u3092 60% \u5411\u4e0a<\/strong><\/li>\n\n\n\n
- API \u30b3\u30b9\u30c8\u3092 30% \u524a\u6e1b<\/strong><\/li>\n<\/ul>\n\n\n\n
1.2 Select\uff08\u9078\u629e\uff09\u2014 \u95a2\u9023\u60c5\u5831\u3060\u3051\u3092\u5229\u7528\u3059\u308b<\/strong><\/h3>\n\n\n\n\u8ab2\u984c:<\/h4>\n\n\n\n
\u4e0d\u8981\u306a\u60c5\u5831\u3092\u4e0e\u3048\u3059\u304e\u308b\u3068\u3001LLM \u306e\u7cbe\u5ea6\u3084\u52b9\u7387\u304c\u4f4e\u4e0b\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
\u89e3\u6c7a\u7b56:<\/h4>\n\n\n\n
RAG<\/em> \u3084 embedding<\/em> \u3092\u4f7f\u3063\u3066\u3001\u95a2\u9023\u5ea6\u306e\u9ad8\u3044\u60c5\u5831\u3084\u30c4\u30fc\u30eb\u3060\u3051\u3092\u9078\u629e\u3002<\/p>\n\n\n\n\u4f8b\uff1aSmartToolSelector<\/strong> \u30bf\u30b9\u30af\u8a18\u8ff0\u306b\u57fa\u3065\u304d\u3001\u6700\u3082\u95a2\u9023\u5ea6\u306e\u9ad8\u3044\u30c4\u30fc\u30eb\u3092\u81ea\u52d5\u9078\u629e\u3002<\/p>\n\n\n\n![\"\"](\"https:\/\/pirago.vn\/wp-content\/uploads\/2025\/09\/image-14.png\")
<\/figure>\n\n\n\n\u52b9\u679c:<\/h4>\n\n\n\n\n- \u30c4\u30fc\u30eb\u9078\u629e\u306e\u7cbe\u5ea6\u304c 31% \u2192 93%<\/strong> \u306b\u5411\u4e0a<\/li>\n\n\n\n
- Document retrieval \u306e\u8aa4\u691c\u51fa\u3092 50% \u6e1b\u5c11<\/strong><\/li>\n\n\n\n
- context \u304c\u8efd\u91cf\u5316\u3057\u51e6\u7406\u901f\u5ea6\u304c 40% \u5411\u4e0a<\/strong><\/li>\n<\/ul>\n\n\n\n
1.3 Compress\uff08\u5727\u7e2e\uff09\u2014 \u5fc5\u8981\u306a tokens \u3060\u3051\u6b8b\u3059<\/strong><\/h3>\n\n\n\n\u8ab2\u984c:<\/h4>\n\n\n\n
API response \u3084 SQL \u7d50\u679c\u306b\u306f\u30ce\u30a4\u30ba\u304c\u591a\u304f\u3001\u305d\u306e\u307e\u307e\u3067\u306f token \u3092\u7121\u99c4\u306b\u6d88\u8cbb\u3002<\/p>\n\n\n\n
\u89e3\u6c7a\u7b56:<\/h4>\n\n\n\n
\u4e0d\u8981\u306a\u90e8\u5206\u3092\u30ab\u30c3\u30c8\u3057\u3001\u8981\u70b9\u3060\u3051\u3092\u5727\u7e2e\u3002<\/p>\n\n\n\n
\u4f8b1\uff1aAPI Response Compression<\/strong> JSON response \u304b\u3089\u91cd\u8981\u30d5\u30a3\u30fc\u30eb\u30c9\u306e\u307f\u62bd\u51fa\u3002<\/p>\n\n\n\n![\"\"](\"https:\/\/pirago.vn\/wp-content\/uploads\/2025\/09\/image-15.png\")
<\/figure>\n\n\n\n\u52b9\u679c:<\/h4>\n\n\n\n\n- tokens \u4f7f\u7528\u91cf\u3092 70% \u524a\u6e1b<\/strong><\/li>\n\n\n\n
- API \u30b3\u30b9\u30c8\u3092 60% \u524a\u6e1b<\/strong><\/li>\n\n\n\n
- \u30ec\u30b9\u30dd\u30f3\u30b9\u901f\u5ea6\u3092 25% \u5411\u4e0a<\/strong><\/li>\n<\/ul>\n\n\n\n
1.4 Isolate\uff08\u5206\u96e2\uff09\u2014 \u554f\u984c\u3092\u5c0f\u3055\u306a context \u306b\u5207\u308a\u5206\u3051\u308b<\/strong><\/h3>\n\n\n\n\u8ab2\u984c:<\/h4>\n\n\n\n
\u8907\u96d1\u306a\u51e6\u7406\u30921\u3064\u306e context \u306b\u62bc\u3057\u8fbc\u3080\u3068\u7cbe\u5ea6\u304c\u4f4e\u4e0b\u3002<\/p>\n\n\n\n
\u89e3\u6c7a\u7b56:<\/h4>\n\n\n\n
\u51e6\u7406\u3092\u30b9\u30c6\u30fc\u30b8\u3054\u3068\u306b\u5206\u3051\u3001\u305d\u308c\u305e\u308c\u72ec\u7acb\u3057\u305f context \u3067\u5b9f\u884c\u3002<\/p>\n\n\n\n
\u4f8b\uff1a<\/strong>UI\u751f\u6210\uff08\u30b9\u30af\u30ea\u30fc\u30f3\u30b7\u30e7\u30c3\u30c8\u304b\u3089\uff09<\/p>\n\n\n\n![\"\"](\"https:\/\/pirago.vn\/wp-content\/uploads\/2025\/09\/image-16.png\")
<\/figure>\n\n\n\n\u52b9\u679c:<\/h4>\n\n\n\n\n- \u5b9f\u88c5\u6210\u529f\u7387 28% \u2192 84%<\/strong><\/li>\n\n\n\n
- Code generation \u306e\u30ed\u30b8\u30c3\u30af\u30a8\u30e9\u30fc 50% \u6e1b\u5c11<\/strong><\/li>\n\n\n\n
- Pattern detection \u306e\u7cbe\u5ea6 40% \u5411\u4e0a<\/strong><\/li>\n<\/ul>\n\n\n\n
2. Prompt Engineering\u3068Context Engineering\u306e\u6bd4\u8f03<\/strong><\/h2>\n\n\n\nPrompt Engineering\uff08\u9759\u7684\uff09<\/strong><\/h3>\n\n\n\n\n- Role Assignment<\/strong>\uff1a\u300cYou are a senior Python developer\u300d<\/li>\n\n\n\n
- Constraints<\/strong>\uff1a\u300c100 words \u4ee5\u5185\u3067\u56de\u7b54\u300d<\/li>\n\n\n\n
- Few-shot Examples<\/strong>\uff1a2\u20133\u500b\u306e\u4f8b\u3092\u63d0\u793a<\/li>\n\n\n\n
- Chain-of-Thought<\/strong>\uff1a\u300cStep by step \u3067\u8003\u3048\u3088\u300d<\/li>\n<\/ul>\n\n\n\n
Context Engineering\uff08\u52d5\u7684\uff09<\/strong><\/h3>\n\n\n\n\n- Memory Management<\/strong>\uff1a\u77ed\u671f\uff08\u4f1a\u8a71\u5c65\u6b74\uff09\uff0b\u9577\u671f\uff08\u30e6\u30fc\u30b6\u30fc\u8a2d\u5b9a\uff09<\/li>\n\n\n\n
- State Management<\/strong>\uff1a\u30ef\u30fc\u30af\u30d5\u30ed\u30fc\u72b6\u614b\u3084\u73fe\u5728\u306e\u30b4\u30fc\u30eb\u3092\u8ffd\u8de1<\/li>\n\n\n\n
- RAG<\/strong>\uff1aKnowledge base \u304b\u3089\u52d5\u7684\u306b\u60c5\u5831\u53d6\u5f97<\/li>\n\n\n\n
- Tool Integration<\/strong>\uff1a\u5916\u90e8 API \u3084 DB \u30af\u30a8\u30ea\u3068\u9023\u643a<\/li>\n<\/ul>\n\n\n\n
3. Cursor AI \u3067\u306e\u5b9f\u8df5\u65b9\u6cd5<\/strong><\/h2>\n\n\n\n![\"\"](\"https:\/\/pirago.vn\/wp-content\/uploads\/2025\/09\/image-17-1024x441.png\")
<\/figure>\n\n\n\n\n- PRD.md<\/a><\/strong>\uff1a\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u8981\u4ef6\u3092\u660e\u78ba\u306b\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u5316<\/li>\n\n\n\n
- auto-generate.mdc<\/strong>\uff1a\u521d\u671f\u30bb\u30c3\u30c8\u30a2\u30c3\u30d7\u3067 docs\/ \u306b\u5fc5\u8981\u8cc7\u6599\u3092\u81ea\u52d5\u751f\u6210<\/li>\n\n\n\n
- workflow.mdc<\/strong>\uff1a\u5168\u3066\u306e\u4f1a\u8a71\u306b\u9069\u7528\u3057\u3001function\u30fb\u69cb\u9020\u30fb\u4fee\u6b63\u5c65\u6b74\u3092\u5171\u6709<\/li>\n\n\n\n
- Implementation<\/strong>\uff1aLLM \u306b\u624b\u9806\u901a\u308a\u5b9f\u88c5\u3055\u305b\u308b<\/li>\n<\/ul>\n\n\n\n
4.\u5b9f\u969b\u306e\u30b1\u30fc\u30b9\u30b9\u30bf\u30c7\u30a3<\/strong><\/h2>\n\n\n\n\u4f8b1: E\u30b3\u30de\u30fc\u30b9\u9867\u5ba2\u5bfe\u5fdc\u30dc\u30c3\u30c8<\/strong><\/h3>\n\n\n\n\n- \u5c0e\u5165\u524d\uff1a45% \u306e\u8cea\u554f\u3092 human agent \u306b\u30a8\u30b9\u30ab\u30ec\u30fc\u30b7\u30e7\u30f3<\/li>\n\n\n\n
- \u5c0e\u5165\u5f8c\uff1a12% \u306b\u6e1b\u5c11\u3001\u5fdc\u7b54\u6642\u9593 60% \u77ed\u7e2e\u3001CS\u30b9\u30b3\u30a2 3.2 \u2192 4.5\/5<\/strong><\/li>\n<\/ul>\n\n\n\n
\u4f8b2: \u30b3\u30fc\u30c9\u30ec\u30d3\u30e5\u30fc\u81ea\u52d5\u5316<\/strong><\/h3>\n\n\n\n\n- 3\u6bb5\u968e Pipeline\uff08syntax \u2192 logic \u2192 best practices\uff09<\/li>\n\n\n\n
- Human reviewer \u304c\u767a\u898b\u3057\u305f\u30d0\u30b0\u306e 85% \u3092\u691c\u51fa<\/strong><\/li>\n\n\n\n
- Review \u6642\u9593\u3092 70% \u77ed\u7e2e<\/strong><\/li>\n\n\n\n
- False positive < 10%<\/strong><\/li>\n<\/ul>\n\n\n\n
5. \u63a8\u5968\u30c4\u30fc\u30eb & \u30ea\u30bd\u30fc\u30b9<\/strong><\/h2>\n\n\n\n\n- LangChain<\/strong>\uff1aContext management framework<\/li>\n\n\n\n
- ChromaDB \/ Pinecone<\/strong>\uff1aVector database for RAG<\/li>\n\n\n\n
- Sentence-Transformers<\/strong>\uff1aEmbedding \u30e2\u30c7\u30eb<\/li>\n\n\n\n
- LlamaIndex<\/strong>\uff1aLLM \u30a2\u30d7\u30ea\u5411\u3051\u30c7\u30fc\u30bf\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af<\/li>\n<\/ul>\n\n\n\n
Monitoring<\/strong><\/p>\n\n\n\n\n- Weights & Biases<\/li>\n\n\n\n
- Langfuse<\/li>\n\n\n\n
- Helicone<\/li>\n<\/ul>\n\n\n\n
\u307e\u3068\u3081<\/strong><\/h2>\n\n\n\nContext Engineering<\/strong> \u3092\u6d3b\u7528\u3059\u308b\u3053\u3068\u3067\u3001AI agent \u306e\u6027\u80fd\u306f\u5927\u5e45\u306b\u5411\u4e0a\u3057\u307e\u3059\u3002<\/p>\n\n\n\n\n- \u52b9\u7387\u6027<\/strong>\uff1a\u8907\u96d1\u30bf\u30b9\u30af\u3067\u6700\u5927 3\u500d\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u5411\u4e0a<\/li>\n\n\n\n
- \u30b3\u30b9\u30c8\u524a\u6e1b<\/strong>\uff1aAPI \u4f7f\u7528\u6599\u3092\u6700\u5927 60% \u524a\u6e1b<\/li>\n\n\n\n
- \u30b9\u30b1\u30fc\u30e9\u30d3\u30ea\u30c6\u30a3<\/strong>\uff1a\u3088\u308a\u8907\u96d1\u306a\u30bf\u30b9\u30af\u3082\u51e6\u7406\u53ef\u80fd<\/li>\n\n\n\n
- \u54c1\u8cea\u7dad\u6301<\/strong>\uff1a\u5927\u898f\u6a21 context \u3067\u3082\u7cbe\u5ea6\u3092\u78ba\u4fdd<\/li>\n<\/ul>\n\n\n\n
\u5927\u5207\u306a\u306e\u306f\u3001\u30e6\u30fc\u30b9\u30b1\u30fc\u30b9\u306b\u5408\u308f\u305b\u305f system design \u3068\u3001\u5b9f\u969b\u306e metrics \u306b\u57fa\u3065\u304f\u7d99\u7d9a\u7684\u306a monitoring & optimization \u3067\u3059\u3002<\/p>\n\n\n\n
\u53c2\u8003\u6587\u732e<\/strong><\/h2>\n\n\n\n\n- Don\u2019t Build Multi-Agents<\/a><\/li>\n\n\n\n
- 12 Factor Agents Guide<\/a><\/li>\n\n\n\n
- LangChain Documentation<\/a><\/li>\n\n\n\n
- O\u2019Reilly: +1 for “context engineering” over “prompt engineering”<\/a><\/li>\n<\/ul>\n\n\n\n
<\/p>",
"protected": false
},
"excerpt": {
"rendered": "
\u306f\u3058\u3081\u306b Context Engineering \u3068\u306f\u3001Large Language Models (LLMs) \u304c\u60c5\u5831\u3092\u51e6\u7406\u3059\u308b\u969b\u306b\u5229\u7528\u3059\u308b\u300c\u30b3\u30f3\u30c6\u30ad\u30b9\u30c8\uff08\u6587\u8108\uff09\u300d\u3092\u7ba1\u7406\u30fb\u6700\u9069\u5316\u3059\u308b\u6280\u8853\u3067\u3059\u3002\u901a\u5e38\u306e Prompt Engineering \u304c\u300c\u30d7\u30ed\u30f3\u30d7\u30c8\uff08\u6307\u793a\u6587\uff09\u306e\u66f8\u304d\u65b9\u300d\u306b\u7126\u70b9\u3092\u5f53\u3066\u3066\u3044\u308b\u306e\u306b\u5bfe\u3057\u3001Context Engineering \u306f\u5165\u529b\uff0f\u51fa\u529b\u5168\u4f53\u3001\u30e1\u30e2\u30ea\u3001AI agent \u306e\u72b6\u614b\u7ba1\u7406\u307e\u3067\u542b\u3081\u3066\u30b3\u30f3\u30c8\u30ed\u30fc\u30eb\u3059\u308b\u70b9\u304c\u7279\u5fb4\u3067\u3059\u3002\u3064\u307e\u308a\u3001\u3059\u3079\u3066\u306f Context Engineering \u3067\u3042\u308b \u3068\u8a00\u3048\u307e\u3059\u3002 1. Context Engineering \u306e4\u3064\u306e\u67f1 1.1 Write\uff08\u66f8\u304d\u51fa\u3057\uff09\u2014 context window \u306e\u5916\u306b\u60c5\u5831\u3092\u4fdd\u5b58\u3059\u308b \u8ab2\u984c: LLMs \u306b\u306f context window \u306e\u4e0a\u9650\u304c\u3042\u308a\u307e\u3059\u3002 \u4f8b\u3048\u3070 GPT-4 \u306f 128k tokens\u3001Claude \u306f 200k tokens\uff08\u7d04 15,000 \u884c\u306e\u30b3\u30fc\u30c9\uff09\u307e\u3067\u3067\u3059\u3002 \u8907\u96d1\u306a\u30bf\u30b9\u30af\u3067\u306f\u91cd\u8981\u60c5\u5831\u304c\u6ea2\u308c\u3066\u3057\u307e\u3046\u3053\u3068\u304c\u3042\u308a\u307e\u3059\u3002 \u89e3\u6c7a\u7b56: \u5916\u90e8\u30b9\u30c8\u30ec\u30fc\u30b8\u306b\u8a08\u753b\u3084\u5c65\u6b74\u3092\u4fdd\u5b58\u3057\u3001\u5fc5\u8981\u306a\u3068\u304d\u306b\u8aad\u307f\u623b\u3057\u307e\u3059\u3002 \u4f8b1\uff1aProject Planner Agent \u5916\u90e8\u30d5\u30a1\u30a4\u30eb\u306b\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u8a08\u753b\u3092\u4fdd\u5b58\u30fb\u53d6\u5f97\u3002 \u4f8b2\uff1aCustomer Support Agent \u904e\u53bb\u306e\u4f1a\u8a71\u5c65\u6b74\u3092 […]<\/p>",
"protected": false
},
"author": 4,
"template": "",
"class_list": [
"post-9919",
"article",
"type-article",
"status-publish",
"hentry"
],
"_links": {
"self": [
{
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/article\/9919",
"targetHints": {
"allow": [
"GET"
]
}
}
],
"collection": [
{
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/article"
}
],
"about": [
{
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/types\/article"
}
],
"author": [
{
"embeddable": true,
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/users\/4"
}
],
"wp:attachment": [
{
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/media?parent=9919"
}
],
"curies": [
{
"name": "wp",
"href": "https:\/\/api.w.org\/{rel}",
"templated": true
}
]
}
}
\u8ab2\u984c:<\/h4>\n\n\n\n
LLMs \u306b\u306f context window<\/em> \u306e\u4e0a\u9650\u304c\u3042\u308a\u307e\u3059\u3002 \u4f8b\u3048\u3070 GPT-4 \u306f 128k tokens\u3001Claude \u306f 200k tokens\uff08\u7d04 15,000 \u884c\u306e\u30b3\u30fc\u30c9\uff09\u307e\u3067\u3067\u3059\u3002 \u8907\u96d1\u306a\u30bf\u30b9\u30af\u3067\u306f\u91cd\u8981\u60c5\u5831\u304c\u6ea2\u308c\u3066\u3057\u307e\u3046\u3053\u3068\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n \u5916\u90e8\u30b9\u30c8\u30ec\u30fc\u30b8\u306b\u8a08\u753b\u3084\u5c65\u6b74\u3092\u4fdd\u5b58\u3057\u3001\u5fc5\u8981\u306a\u3068\u304d\u306b\u8aad\u307f\u623b\u3057\u307e\u3059\u3002<\/p>\n\n\n\n \u4f8b1\uff1aProject Planner Agent<\/strong> \u5916\u90e8\u30d5\u30a1\u30a4\u30eb\u306b\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u8a08\u753b\u3092\u4fdd\u5b58\u30fb\u53d6\u5f97\u3002<\/p>\n\n\n\n \u4f8b2\uff1aCustomer Support Agent<\/strong> \u904e\u53bb\u306e\u4f1a\u8a71\u5c65\u6b74\u3092 SQLite \u306b\u4fdd\u5b58\u3057\u3001\u30e6\u30fc\u30b6\u30fc\u3054\u3068\u306e\u6587\u8108\u3092\u547c\u3073\u51fa\u3059\u3002<\/p>\n\n\n\n \u4e0d\u8981\u306a\u60c5\u5831\u3092\u4e0e\u3048\u3059\u304e\u308b\u3068\u3001LLM \u306e\u7cbe\u5ea6\u3084\u52b9\u7387\u304c\u4f4e\u4e0b\u3057\u307e\u3059\u3002<\/p>\n\n\n\n RAG<\/em> \u3084 embedding<\/em> \u3092\u4f7f\u3063\u3066\u3001\u95a2\u9023\u5ea6\u306e\u9ad8\u3044\u60c5\u5831\u3084\u30c4\u30fc\u30eb\u3060\u3051\u3092\u9078\u629e\u3002<\/p>\n\n\n\n \u4f8b\uff1aSmartToolSelector<\/strong> \u30bf\u30b9\u30af\u8a18\u8ff0\u306b\u57fa\u3065\u304d\u3001\u6700\u3082\u95a2\u9023\u5ea6\u306e\u9ad8\u3044\u30c4\u30fc\u30eb\u3092\u81ea\u52d5\u9078\u629e\u3002<\/p>\n\n\n\n API response \u3084 SQL \u7d50\u679c\u306b\u306f\u30ce\u30a4\u30ba\u304c\u591a\u304f\u3001\u305d\u306e\u307e\u307e\u3067\u306f token \u3092\u7121\u99c4\u306b\u6d88\u8cbb\u3002<\/p>\n\n\n\n \u4e0d\u8981\u306a\u90e8\u5206\u3092\u30ab\u30c3\u30c8\u3057\u3001\u8981\u70b9\u3060\u3051\u3092\u5727\u7e2e\u3002<\/p>\n\n\n\n \u4f8b1\uff1aAPI Response Compression<\/strong> JSON response \u304b\u3089\u91cd\u8981\u30d5\u30a3\u30fc\u30eb\u30c9\u306e\u307f\u62bd\u51fa\u3002<\/p>\n\n\n\n \u8907\u96d1\u306a\u51e6\u7406\u30921\u3064\u306e context \u306b\u62bc\u3057\u8fbc\u3080\u3068\u7cbe\u5ea6\u304c\u4f4e\u4e0b\u3002<\/p>\n\n\n\n \u51e6\u7406\u3092\u30b9\u30c6\u30fc\u30b8\u3054\u3068\u306b\u5206\u3051\u3001\u305d\u308c\u305e\u308c\u72ec\u7acb\u3057\u305f context \u3067\u5b9f\u884c\u3002<\/p>\n\n\n\n \u4f8b\uff1a<\/strong>UI\u751f\u6210\uff08\u30b9\u30af\u30ea\u30fc\u30f3\u30b7\u30e7\u30c3\u30c8\u304b\u3089\uff09<\/p>\n\n\n\n Monitoring<\/strong><\/p>\n\n\n\n Context Engineering<\/strong> \u3092\u6d3b\u7528\u3059\u308b\u3053\u3068\u3067\u3001AI agent \u306e\u6027\u80fd\u306f\u5927\u5e45\u306b\u5411\u4e0a\u3057\u307e\u3059\u3002<\/p>\n\n\n\n \u5927\u5207\u306a\u306e\u306f\u3001\u30e6\u30fc\u30b9\u30b1\u30fc\u30b9\u306b\u5408\u308f\u305b\u305f system design \u3068\u3001\u5b9f\u969b\u306e metrics \u306b\u57fa\u3065\u304f\u7d99\u7d9a\u7684\u306a monitoring & optimization \u3067\u3059\u3002<\/p>\n\n\n\n <\/p>",
"protected": false
},
"excerpt": {
"rendered": " \u306f\u3058\u3081\u306b Context Engineering \u3068\u306f\u3001Large Language Models (LLMs) \u304c\u60c5\u5831\u3092\u51e6\u7406\u3059\u308b\u969b\u306b\u5229\u7528\u3059\u308b\u300c\u30b3\u30f3\u30c6\u30ad\u30b9\u30c8\uff08\u6587\u8108\uff09\u300d\u3092\u7ba1\u7406\u30fb\u6700\u9069\u5316\u3059\u308b\u6280\u8853\u3067\u3059\u3002\u901a\u5e38\u306e Prompt Engineering \u304c\u300c\u30d7\u30ed\u30f3\u30d7\u30c8\uff08\u6307\u793a\u6587\uff09\u306e\u66f8\u304d\u65b9\u300d\u306b\u7126\u70b9\u3092\u5f53\u3066\u3066\u3044\u308b\u306e\u306b\u5bfe\u3057\u3001Context Engineering \u306f\u5165\u529b\uff0f\u51fa\u529b\u5168\u4f53\u3001\u30e1\u30e2\u30ea\u3001AI agent \u306e\u72b6\u614b\u7ba1\u7406\u307e\u3067\u542b\u3081\u3066\u30b3\u30f3\u30c8\u30ed\u30fc\u30eb\u3059\u308b\u70b9\u304c\u7279\u5fb4\u3067\u3059\u3002\u3064\u307e\u308a\u3001\u3059\u3079\u3066\u306f Context Engineering \u3067\u3042\u308b \u3068\u8a00\u3048\u307e\u3059\u3002 1. Context Engineering \u306e4\u3064\u306e\u67f1 1.1 Write\uff08\u66f8\u304d\u51fa\u3057\uff09\u2014 context window \u306e\u5916\u306b\u60c5\u5831\u3092\u4fdd\u5b58\u3059\u308b \u8ab2\u984c: LLMs \u306b\u306f context window \u306e\u4e0a\u9650\u304c\u3042\u308a\u307e\u3059\u3002 \u4f8b\u3048\u3070 GPT-4 \u306f 128k tokens\u3001Claude \u306f 200k tokens\uff08\u7d04 15,000 \u884c\u306e\u30b3\u30fc\u30c9\uff09\u307e\u3067\u3067\u3059\u3002 \u8907\u96d1\u306a\u30bf\u30b9\u30af\u3067\u306f\u91cd\u8981\u60c5\u5831\u304c\u6ea2\u308c\u3066\u3057\u307e\u3046\u3053\u3068\u304c\u3042\u308a\u307e\u3059\u3002 \u89e3\u6c7a\u7b56: \u5916\u90e8\u30b9\u30c8\u30ec\u30fc\u30b8\u306b\u8a08\u753b\u3084\u5c65\u6b74\u3092\u4fdd\u5b58\u3057\u3001\u5fc5\u8981\u306a\u3068\u304d\u306b\u8aad\u307f\u623b\u3057\u307e\u3059\u3002 \u4f8b1\uff1aProject Planner Agent \u5916\u90e8\u30d5\u30a1\u30a4\u30eb\u306b\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u8a08\u753b\u3092\u4fdd\u5b58\u30fb\u53d6\u5f97\u3002 \u4f8b2\uff1aCustomer Support Agent \u904e\u53bb\u306e\u4f1a\u8a71\u5c65\u6b74\u3092 […]<\/p>",
"protected": false
},
"author": 4,
"template": "",
"class_list": [
"post-9919",
"article",
"type-article",
"status-publish",
"hentry"
],
"_links": {
"self": [
{
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/article\/9919",
"targetHints": {
"allow": [
"GET"
]
}
}
],
"collection": [
{
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/article"
}
],
"about": [
{
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/types\/article"
}
],
"author": [
{
"embeddable": true,
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/users\/4"
}
],
"wp:attachment": [
{
"href": "https:\/\/pirago.vn\/vn\/wp-json\/wp\/v2\/media?parent=9919"
}
],
"curies": [
{
"name": "wp",
"href": "https:\/\/api.w.org\/{rel}",
"templated": true
}
]
}
}\u89e3\u6c7a\u7b56:<\/h4>\n\n\n\n
<\/figure>\n\n\n\n<\/figure>\n\n\n\n\u52b9\u679c:<\/h4>\n\n\n\n
\n
1.2 Select\uff08\u9078\u629e\uff09\u2014 \u95a2\u9023\u60c5\u5831\u3060\u3051\u3092\u5229\u7528\u3059\u308b<\/strong><\/h3>\n\n\n\n
\u8ab2\u984c:<\/h4>\n\n\n\n
\u89e3\u6c7a\u7b56:<\/h4>\n\n\n\n
<\/figure>\n\n\n\n\u52b9\u679c:<\/h4>\n\n\n\n
\n
1.3 Compress\uff08\u5727\u7e2e\uff09\u2014 \u5fc5\u8981\u306a tokens \u3060\u3051\u6b8b\u3059<\/strong><\/h3>\n\n\n\n
\u8ab2\u984c:<\/h4>\n\n\n\n
\u89e3\u6c7a\u7b56:<\/h4>\n\n\n\n
<\/figure>\n\n\n\n\u52b9\u679c:<\/h4>\n\n\n\n
\n
1.4 Isolate\uff08\u5206\u96e2\uff09\u2014 \u554f\u984c\u3092\u5c0f\u3055\u306a context \u306b\u5207\u308a\u5206\u3051\u308b<\/strong><\/h3>\n\n\n\n
\u8ab2\u984c:<\/h4>\n\n\n\n
\u89e3\u6c7a\u7b56:<\/h4>\n\n\n\n
<\/figure>\n\n\n\n\u52b9\u679c:<\/h4>\n\n\n\n
\n
2. Prompt Engineering\u3068Context Engineering\u306e\u6bd4\u8f03<\/strong><\/h2>\n\n\n\n
Prompt Engineering\uff08\u9759\u7684\uff09<\/strong><\/h3>\n\n\n\n
\n
Context Engineering\uff08\u52d5\u7684\uff09<\/strong><\/h3>\n\n\n\n
\n
3. Cursor AI \u3067\u306e\u5b9f\u8df5\u65b9\u6cd5<\/strong><\/h2>\n\n\n\n
<\/figure>\n\n\n\n\n
4.\u5b9f\u969b\u306e\u30b1\u30fc\u30b9\u30b9\u30bf\u30c7\u30a3<\/strong><\/h2>\n\n\n\n
\u4f8b1: E\u30b3\u30de\u30fc\u30b9\u9867\u5ba2\u5bfe\u5fdc\u30dc\u30c3\u30c8<\/strong><\/h3>\n\n\n\n
\n
\u4f8b2: \u30b3\u30fc\u30c9\u30ec\u30d3\u30e5\u30fc\u81ea\u52d5\u5316<\/strong><\/h3>\n\n\n\n
\n
5. \u63a8\u5968\u30c4\u30fc\u30eb & \u30ea\u30bd\u30fc\u30b9<\/strong><\/h2>\n\n\n\n
\n
\n
\u307e\u3068\u3081<\/strong><\/h2>\n\n\n\n
\n
\u53c2\u8003\u6587\u732e<\/strong><\/h2>\n\n\n\n
\n