Repository navigation
Expand file tree
/
Copy pathindex.html
More file actions
1590 lines (1522 loc) · 84.4 KB
/
Copy pathindex.html
File metadata and controls
1590 lines (1522 loc) · 84.4 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="ko">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta name="guide-version" content="2026.04">
<title>AI Agent 완벽 가이드 - From Zero to Multi-Agent</title>
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700;800;900&family=JetBrains+Mono:wght@400;500;600&family=Noto+Sans+KR:wght@300;400;500;600;700;800;900&display=swap" rel="stylesheet">
<link rel="stylesheet" href="common/base.css">
<link rel="stylesheet" href="style.css">
<!-- Google Analytics (GA4) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4L031NBC3Z"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-4L031NBC3Z');
</script>
</head>
<body>
<!-- Progress Bar -->
<div class="reading-progress" id="readingProgress"></div>
<!-- Navigation -->
<nav class="nav" id="nav">
<div class="nav-inner">
<a href="#hero" class="nav-logo">
<span class="nav-logo-icon">A</span>
<span>Agent Guide</span>
</a>
<div class="nav-links" id="navLinks">
<a href="#overview" class="nav-link">Overview</a>
<a href="#levels" class="nav-link">Levels</a>
<a href="#architecture" class="nav-link">Architecture</a>
<a href="#protocols" class="nav-link">Protocols</a>
<a href="#practice" class="nav-link">Practice</a>
<a href="#papers" class="nav-link">Papers</a>
</div>
<a href="en/" class="lang-toggle" title="Read in English">EN</a>
<button class="nav-toggle" id="navToggle" aria-label="Toggle menu">
<span></span><span></span><span></span>
</button>
</div>
</nav>
<!-- Hero -->
<section class="hero" id="hero">
<div class="hero-bg">
<div class="hero-grid"></div>
<div class="hero-glow hero-glow-1"></div>
<div class="hero-glow hero-glow-2"></div>
</div>
<div class="container hero-content">
<div class="hero-badge">2026 Complete Guide</div>
<h1 class="hero-title">
AI Agent<br>
<span class="hero-gradient">완벽 가이드</span>
</h1>
<p class="hero-subtitle">
단순 LLM 호출부터 멀티 에이전트 시스템까지<br>
7단계로 이해하는 AI Agent의 모든 것
</p>
<div class="hero-stats">
<div class="hero-stat">
<span class="hero-stat-number">7</span>
<span class="hero-stat-label">Agent Levels</span>
</div>
<div class="hero-stat">
<span class="hero-stat-number">10+</span>
<span class="hero-stat-label">핵심 논문</span>
</div>
<div class="hero-stat">
<span class="hero-stat-number">6</span>
<span class="hero-stat-label">주요 프레임워크</span>
</div>
</div>
<div class="hero-actions">
<a href="#levels" class="btn btn-primary">레벨별 가이드 보기</a>
<a href="#overview" class="btn btn-secondary">Overview 먼저 보기</a>
</div>
</div>
<div class="hero-scroll">
<div class="hero-scroll-line"></div>
</div>
</section>
<!-- Overview Section -->
<section class="section" id="overview">
<div class="container">
<div class="section-header">
<span class="section-label">Overview</span>
<h2 class="section-title">AI Agent란 무엇인가?</h2>
<p class="section-desc">Lilian Weng(OpenAI)의 정의에 따르면, AI Agent는 네 가지 핵심 요소의 조합입니다</p>
</div>
<!-- Agent = LLM + Memory + Planning + Tools -->
<div class="formula-card">
<div class="formula">
<span class="formula-item formula-agent">Agent</span>
<span class="formula-op">=</span>
<span class="formula-item formula-llm">LLM</span>
<span class="formula-op">+</span>
<span class="formula-item formula-memory">Memory</span>
<span class="formula-op">+</span>
<span class="formula-item formula-planning">Planning</span>
<span class="formula-op">+</span>
<span class="formula-item formula-tools">Tools</span>
</div>
<p class="formula-source">Source: Lilian Weng, "LLM Powered Autonomous Agents" (June 2023)</p>
</div>
<!-- 4 Components -->
<div class="components-grid">
<div class="component-card" data-color="blue">
<div class="component-icon">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M12 2a7 7 0 0 1 7 7c0 3-2 5.5-4 7.5L12 22l-3-5.5C7 14.5 5 12 5 9a7 7 0 0 1 7-7z"/><circle cx="12" cy="9" r="2"/></svg>
</div>
<h3>LLM (두뇌)</h3>
<p>추론과 의사결정의 핵심 엔진. 자연어를 이해하고, 계획을 세우고, 도구 사용을 결정합니다.</p>
<div class="component-tag">Core Engine</div>
</div>
<div class="component-card" data-color="purple">
<div class="component-icon">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><rect x="2" y="6" width="20" height="12" rx="2"/><path d="M6 10h.01M10 10h.01M14 10h.01M6 14h12"/></svg>
</div>
<h3>Memory (기억)</h3>
<p>단기 기억(컨텍스트 윈도우)과 장기 기억(벡터 DB). 경험을 축적하고 과거를 참조합니다.</p>
<div class="component-tag">State Management</div>
</div>
<div class="component-card" data-color="green">
<div class="component-icon">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M9 5H7a2 2 0 0 0-2 2v12a2 2 0 0 0 2 2h10a2 2 0 0 0 2-2V7a2 2 0 0 0-2-2h-2"/><rect x="9" y="3" width="6" height="4" rx="1"/><path d="M9 12l2 2 4-4"/></svg>
</div>
<h3>Planning (계획)</h3>
<p>작업 분해(Task Decomposition)와 자기 반성(Reflection). 복잡한 목표를 실행 가능한 단계로 쪼갭니다.</p>
<div class="component-tag">Strategy</div>
</div>
<div class="component-card" data-color="orange">
<div class="component-icon">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M14.7 6.3a1 1 0 0 0 0 1.4l1.6 1.6a1 1 0 0 0 1.4 0l3.77-3.77a6 6 0 0 1-7.94 7.94l-6.91 6.91a2.12 2.12 0 0 1-3-3l6.91-6.91a6 6 0 0 1 7.94-7.94l-3.76 3.76z"/></svg>
</div>
<h3>Tools (도구)</h3>
<p>외부 API, 검색 엔진, 코드 실행기 등. LLM의 능력을 실제 세계로 확장합니다.</p>
<div class="component-tag">External Actions</div>
</div>
</div>
<!-- Workflow vs Agent -->
<div class="comparison-section">
<h3 class="subsection-title">Workflow vs Agent: 핵심 구분</h3>
<p class="subsection-desc">Anthropic의 "Building Effective Agents"(2024)에서는 Workflow와 Agent를 명확히 구분합니다</p>
<div class="comparison-table">
<div class="comparison-col comparison-workflow">
<div class="comparison-header">
<h4>Workflow</h4>
<span class="comparison-badge">Deterministic</span>
</div>
<ul>
<li>실행 흐름이 <strong>코드로 미리 정의</strong>됨</li>
<li>같은 입력 = 같은 경로</li>
<li>예측 가능하고 디버깅이 쉬움</li>
<li>대부분의 비즈니스 문제에 적합</li>
<li>비용이 예측 가능</li>
</ul>
<div class="comparison-example">
<strong>예시:</strong> 문서 번역 파이프라인, 이메일 분류 시스템
</div>
</div>
<div class="comparison-vs">vs</div>
<div class="comparison-col comparison-agent">
<div class="comparison-header">
<h4>Agent</h4>
<span class="comparison-badge">Dynamic</span>
</div>
<ul>
<li>실행 흐름을 <strong>LLM이 동적으로 결정</strong></li>
<li>같은 입력이라도 다른 경로 가능</li>
<li>관찰(Observability) 도구 필요</li>
<li>Open-ended 문제에 강함</li>
<li>비용이 가변적</li>
</ul>
<div class="comparison-example">
<strong>예시:</strong> 코드 디버깅 에이전트, 리서치 에이전트
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Levels Section -->
<section class="section section-dark" id="levels">
<div class="container">
<div class="section-header">
<span class="section-label">Levels 0-6</span>
<h2 class="section-title">Agent 성숙도 7단계</h2>
<p class="section-desc">단순 LLM 호출부터 멀티 에이전트 시스템까지, 각 레벨의 구조와 특징을 살펴봅니다</p>
</div>
<!-- Cross-link to patterns guide -->
<div style="max-width: 860px; margin: 0 auto 40px; padding: 20px 24px; background: linear-gradient(135deg, rgba(251,146,60,0.06), rgba(244,114,182,0.06)); border: 1px solid rgba(251,146,60,0.25); border-radius: 14px; display: flex; align-items: center; gap: 18px; flex-wrap: wrap;">
<div style="font-size: 2rem; line-height: 1;">🧩</div>
<div style="flex: 1; min-width: 240px;">
<div style="font-size: 0.78rem; font-weight: 700; color: #fb923c; text-transform: uppercase; letter-spacing: 0.06em; margin-bottom: 4px;">함께 보기</div>
<div style="font-size: 0.95rem; color: var(--text); line-height: 1.55;">
<strong>어떤 패턴을 내 상황에 써야 할지</strong> 선택이 고민이라면 —
패턴별 실전 예시·비용/레이턴시 비교·조합 레시피가 담긴
<a href="https://desty.github.io/agentic-patterns-guide/" style="color: #fb923c; font-weight: 700; text-decoration: none; border-bottom: 1.5px solid rgba(251,146,60,0.4);">Agentic AI 패턴 가이드</a>를 보세요.
</div>
</div>
<a href="https://desty.github.io/agentic-patterns-guide/" style="padding: 10px 18px; background: rgba(251,146,60,0.15); border: 1px solid rgba(251,146,60,0.4); border-radius: 999px; color: #fb923c; font-weight: 700; font-size: 0.88rem; text-decoration: none; white-space: nowrap;">패턴 가이드 →</a>
</div>
<!-- Level Progress Bar -->
<div class="level-progress">
<div class="level-progress-track">
<div class="level-progress-fill"></div>
</div>
<div class="level-progress-labels">
<span>Simple</span>
<span>Autonomous</span>
<span>Multi-Agent</span>
</div>
</div>
<!-- Level 0 -->
<div class="level-card" id="level-0" data-level="0">
<div class="level-header">
<div class="level-badge">L0</div>
<div class="level-info">
<h3>Simple LLM Call</h3>
<div class="level-tags">
<span class="tag tag-red">No Tools</span>
<span class="tag tag-red">No Memory</span>
<span class="tag tag-gray">Single Turn</span>
</div>
</div>
<div class="level-autonomy">
<span class="autonomy-label">자율성</span>
<div class="autonomy-bar"><div class="autonomy-fill" style="width:5%"></div></div>
</div>
</div>
<div class="level-body">
<div class="level-desc">
<p>가장 기본적인 형태입니다. 프롬프트를 넣으면 응답이 나오는 단순 호출로, LLM의 학습된 지식만으로 답변합니다. 외부 도구 접근이 없어 <strong>할루시네이션 리스크가 가장 높습니다.</strong></p>
</div>
<!-- Architecture Diagram -->
<div class="arch-diagram">
<div class="arch-flow">
<div class="arch-node arch-user">
<div class="arch-node-icon">U</div>
<span>User</span>
</div>
<div class="arch-arrow">
<span>Prompt</span>
</div>
<div class="arch-node arch-llm">
<div class="arch-node-icon">L</div>
<span>LLM</span>
</div>
<div class="arch-arrow">
<span>Response</span>
</div>
<div class="arch-node arch-output">
<div class="arch-node-icon">R</div>
<span>Output</span>
</div>
</div>
</div>
<div class="level-details">
<div class="detail-col">
<h4>사용 예시</h4>
<ul>
<li>"이메일 초안 써줘"</li>
<li>"이 코드 리뷰해줘"</li>
<li>"마케팅 카피 만들어줘"</li>
</ul>
</div>
<div class="detail-col">
<h4>한계</h4>
<ul>
<li>최신 정보 접근 불가</li>
<li>사내 데이터 참조 불가</li>
<li>정보를 지어낼 수 있음</li>
</ul>
</div>
<div class="detail-col">
<h4>관련 기술</h4>
<ul>
<li>Chain-of-Thought (Wei et al., 2022)</li>
<li>Zero-shot / Few-shot Prompting</li>
<li>기본 ChatGPT / Claude 호출</li>
</ul>
</div>
</div>
</div>
</div>
<!-- Level 1 -->
<div class="level-card" id="level-1" data-level="1">
<div class="level-header">
<div class="level-badge">L1</div>
<div class="level-info">
<h3>Augmented LLM (Tool Use)</h3>
<div class="level-tags">
<span class="tag tag-blue">Function Calling</span>
<span class="tag tag-blue">RAG</span>
<span class="tag tag-gray">Single Cycle</span>
</div>
</div>
<div class="level-autonomy">
<span class="autonomy-label">자율성</span>
<div class="autonomy-bar"><div class="autonomy-fill" style="width:20%"></div></div>
</div>
</div>
<div class="level-body">
<div class="level-desc">
<p>LLM이 필요할 때 <strong>외부 도구를 한 번 호출</strong>할 수 있습니다. 도구를 쓸지 말지를 LLM이 판단하고, 결과를 받아 최종 응답을 생성합니다. 하지만 <strong>한 번의 사이클로 끝나며</strong>, 결과가 부족해도 재시도하지 않습니다.</p>
</div>
<div class="arch-diagram">
<div class="arch-flow">
<div class="arch-node arch-user">
<div class="arch-node-icon">U</div>
<span>User</span>
</div>
<div class="arch-arrow"><span>Query</span></div>
<div class="arch-node arch-llm arch-node-lg">
<div class="arch-node-icon">L</div>
<span>LLM</span>
<div class="arch-sub-nodes">
<div class="arch-tool-call">
<span class="arch-tool-label">Tool Call</span>
<div class="arch-mini-arrow"></div>
<div class="arch-node-mini arch-tool">API / DB / Search</div>
</div>
</div>
</div>
<div class="arch-arrow"><span>Response</span></div>
<div class="arch-node arch-output">
<div class="arch-node-icon">R</div>
<span>Output</span>
</div>
</div>
</div>
<div class="level-details">
<div class="detail-col">
<h4>도구 유형</h4>
<ul>
<li><strong>검색:</strong> RAG, 웹 검색</li>
<li><strong>API:</strong> 날씨, 주가, DB 쿼리</li>
<li><strong>실행:</strong> 코드 인터프리터, 계산기</li>
</ul>
</div>
<div class="detail-col">
<h4>사용 예시</h4>
<ul>
<li>"오늘 서울 날씨 알려줘" → weather API</li>
<li>"Q3 매출 데이터 찾아줘" → DB 쿼리</li>
<li>Naive RAG: 벡터 검색 → 답변 생성</li>
</ul>
</div>
<div class="detail-col">
<h4>관련 기술</h4>
<ul>
<li>Toolformer (Schick et al., 2023)</li>
<li>MRKL Systems (Karpas et al., 2022)</li>
<li>Function Calling / Structured Output</li>
</ul>
</div>
</div>
</div>
</div>
<!-- Level 2 -->
<div class="level-card" id="level-2" data-level="2">
<div class="level-header">
<div class="level-badge">L2</div>
<div class="level-info">
<h3>Chained / Sequential Agent</h3>
<div class="level-tags">
<span class="tag tag-green">Pipeline</span>
<span class="tag tag-green">Deterministic</span>
<span class="tag tag-gray">Multi-Step</span>
</div>
</div>
<div class="level-autonomy">
<span class="autonomy-label">자율성</span>
<div class="autonomy-bar"><div class="autonomy-fill" style="width:30%"></div></div>
</div>
</div>
<div class="level-body">
<div class="level-desc">
<p>여러 단계를 <strong>미리 정의된 순서대로</strong> 실행합니다. 앞 단계의 출력이 뒷 단계의 입력이 됩니다. Gate(검증 단계)를 넣어 품질을 체크할 수 있지만, 전체 흐름은 코드로 고정되어 있습니다.</p>
</div>
<div class="arch-diagram">
<div class="arch-pipeline">
<div class="pipe-node">
<div class="pipe-icon">1</div>
<span>분석</span>
</div>
<div class="pipe-connector">
<div class="pipe-line"></div>
<div class="pipe-arrow-head"></div>
</div>
<div class="pipe-node">
<div class="pipe-icon">2</div>
<span>처리</span>
</div>
<div class="pipe-connector">
<div class="pipe-line"></div>
<div class="pipe-arrow-head"></div>
</div>
<div class="pipe-gate">
<div class="pipe-icon">G</div>
<span>검증</span>
</div>
<div class="pipe-connector">
<div class="pipe-line"></div>
<div class="pipe-arrow-head"></div>
</div>
<div class="pipe-node">
<div class="pipe-icon">3</div>
<span>출력</span>
</div>
</div>
</div>
<div class="example-cards">
<div class="example-card">
<h5>문서 번역 파이프라인</h5>
<div class="pipeline-steps">
<span>원문 분석</span>
<span class="step-arrow">→</span>
<span>초벌 번역</span>
<span class="step-arrow">→</span>
<span>용어 검증</span>
<span class="step-arrow">→</span>
<span>최종 다듬기</span>
</div>
</div>
<div class="example-card">
<h5>코드 생성 파이프라인</h5>
<div class="pipeline-steps">
<span>요구사항</span>
<span class="step-arrow">→</span>
<span>코드 생성</span>
<span class="step-arrow">→</span>
<span>린트/테스트</span>
<span class="step-arrow">→</span>
<span>리뷰/수정</span>
</div>
</div>
</div>
<div class="level-details">
<div class="detail-col">
<h4>특징</h4>
<ul>
<li>실행 순서가 코드로 고정 (deterministic)</li>
<li>각 단계에서 LLM이 동작하지만 전체 흐름은 LLM이 결정하지 않음</li>
<li>Latency = 각 단계의 합산</li>
</ul>
</div>
<div class="detail-col">
<h4>한계</h4>
<ul>
<li>분기(branching)가 없음</li>
<li>간단한 요청도 전체 파이프라인 통과 필요</li>
<li>한 단계 실패 시 전체 중단</li>
</ul>
</div>
</div>
</div>
</div>
<!-- Level 3 -->
<div class="level-card" id="level-3" data-level="3">
<div class="level-header">
<div class="level-badge">L3</div>
<div class="level-info">
<h3>Router / Branching Agent</h3>
<div class="level-tags">
<span class="tag tag-yellow">Dynamic Routing</span>
<span class="tag tag-yellow">Classifier</span>
<span class="tag tag-gray">Parallel</span>
</div>
</div>
<div class="level-autonomy">
<span class="autonomy-label">자율성</span>
<div class="autonomy-bar"><div class="autonomy-fill" style="width:45%"></div></div>
</div>
</div>
<div class="level-body">
<div class="level-desc">
<p>입력을 분석해서 <strong>적절한 경로로 분기</strong>합니다. LLM이 분류기(classifier) 역할을 수행하며, 각 브랜치는 독립적인 워크플로우입니다. 리소스 효율적이고 병렬 실행도 가능합니다.</p>
</div>
<div class="arch-diagram">
<div class="arch-router">
<div class="router-input">
<div class="arch-node arch-user">
<div class="arch-node-icon">U</div>
<span>Input</span>
</div>
</div>
<div class="router-center">
<div class="arch-node arch-llm">
<div class="arch-node-icon">R</div>
<span>Router</span>
</div>
</div>
<div class="router-branches">
<div class="branch branch-a">
<div class="branch-line"></div>
<div class="branch-node">
<span class="branch-label">Simple</span>
<span class="branch-desc">직접 답변</span>
</div>
</div>
<div class="branch branch-b">
<div class="branch-line"></div>
<div class="branch-node">
<span class="branch-label">Data</span>
<span class="branch-desc">검색 파이프라인</span>
</div>
</div>
<div class="branch branch-c">
<div class="branch-line"></div>
<div class="branch-node">
<span class="branch-label">Code</span>
<span class="branch-desc">코드 생성</span>
</div>
</div>
</div>
</div>
</div>
<div class="example-cards">
<div class="example-card">
<h5>고객 지원 시스템</h5>
<div class="router-example">
<div class="router-ex-input">고객 메시지</div>
<div class="router-ex-branches">
<div>"환불 요청" → 환불 처리 워크플로우</div>
<div>"기술 문의" → 기술 지원 워크플로우</div>
<div>"일반 문의" → FAQ 기반 응답</div>
</div>
</div>
</div>
</div>
</div>
</div>
<!-- Level 4 - Key level -->
<div class="level-card level-card-highlight" id="level-4" data-level="4">
<div class="level-header">
<div class="level-badge">L4</div>
<div class="level-info">
<h3>ReAct / Loop Agent</h3>
<div class="level-tags">
<span class="tag tag-purple">True Agent</span>
<span class="tag tag-purple">Self-Correction</span>
<span class="tag tag-purple">Autonomous Loop</span>
</div>
</div>
<div class="level-autonomy">
<span class="autonomy-label">자율성</span>
<div class="autonomy-bar"><div class="autonomy-fill" style="width:65%"></div></div>
</div>
</div>
<div class="level-milestone">여기서부터 진정한 Agent입니다</div>
<div class="level-body">
<div class="level-desc">
<p>LLM이 <strong>루프를 돌면서 스스로 판단하고, 도구를 선택하고, 결과를 평가하고, 필요하면 다시 시도</strong>합니다. ReAct(Reasoning + Acting) 패턴이 핵심이며, 종료 조건도 LLM이 판단합니다.</p>
</div>
<div class="arch-diagram">
<div class="arch-react-loop">
<div class="react-cycle">
<div class="react-loop-back">
<span class="loop-label">반복</span>
</div>
<div class="react-nodes-row">
<div class="react-node react-think">
<div class="react-icon">T</div>
<span>Thought</span>
<p>상황 분석 & 판단</p>
</div>
<div class="react-arrow"></div>
<div class="react-node react-act">
<div class="react-icon">A</div>
<span>Action</span>
<p>도구 선택 & 실행</p>
</div>
<div class="react-arrow"></div>
<div class="react-node react-observe">
<div class="react-icon">O</div>
<span>Observe</span>
<p>결과 확인 & 평가</p>
</div>
</div>
</div>
<div class="react-exit">
<div class="react-exit-arrow"></div>
<div class="react-node react-answer">
<div class="react-icon">!</div>
<span>Answer</span>
</div>
</div>
</div>
</div>
<!-- ReAct Example -->
<div class="react-example">
<h4>ReAct 실행 예시: 리서치 에이전트</h4>
<div class="react-trace">
<div class="trace-step trace-thought">
<div class="trace-label">Thought</div>
<div class="trace-content">사용자가 2024년 한국 GDP를 묻고 있다. 최신 데이터니까 검색이 필요하다.</div>
</div>
<div class="trace-step trace-action">
<div class="trace-label">Action</div>
<div class="trace-content"><code>web_search("2024 한국 GDP")</code></div>
</div>
<div class="trace-step trace-observe">
<div class="trace-label">Observe</div>
<div class="trace-content">검색 결과가 2023년 데이터뿐이다. 더 최신 쿼리가 필요.</div>
</div>
<div class="trace-step trace-thought">
<div class="trace-label">Thought</div>
<div class="trace-content">쿼리를 변경해서 재검색하자.</div>
</div>
<div class="trace-step trace-action">
<div class="trace-label">Action</div>
<div class="trace-content"><code>web_search("South Korea GDP 2024 IMF estimate")</code></div>
</div>
<div class="trace-step trace-observe">
<div class="trace-label">Observe</div>
<div class="trace-content">IMF 기준 1.7조 달러라는 결과 확인. 신뢰할 만한 출처.</div>
</div>
<div class="trace-step trace-answer">
<div class="trace-label">Answer</div>
<div class="trace-content">2024년 한국 GDP는 IMF 추정 기준 약 1.7조 달러입니다...</div>
</div>
</div>
</div>
<div class="level-details">
<div class="detail-col">
<h4>핵심 특징</h4>
<ul>
<li>LLM이 실행 흐름을 제어</li>
<li>자기 수정(self-correction) 가능</li>
<li>종료 조건을 LLM이 판단</li>
<li>max_iterations 설정 필수</li>
</ul>
</div>
<div class="detail-col">
<h4>관련 논문</h4>
<ul>
<li>ReAct (Yao et al., 2022) - ICLR 2023</li>
<li>Reflexion (Shinn et al., 2023)</li>
<li>Tree of Thoughts (Yao et al., 2023)</li>
</ul>
</div>
</div>
</div>
</div>
<!-- Level 5 -->
<div class="level-card" id="level-5" data-level="5">
<div class="level-header">
<div class="level-badge">L5</div>
<div class="level-info">
<h3>Planning Agent</h3>
<div class="level-tags">
<span class="tag tag-cyan">Task Decomposition</span>
<span class="tag tag-cyan">Adaptive Replanning</span>
<span class="tag tag-cyan">Reflection</span>
</div>
</div>
<div class="level-autonomy">
<span class="autonomy-label">자율성</span>
<div class="autonomy-bar"><div class="autonomy-fill" style="width:80%"></div></div>
</div>
</div>
<div class="level-body">
<div class="level-desc">
<p>실행 전에 먼저 <strong>전체 계획을 세우고</strong>, 계획에 따라 단계별로 실행하며, 상황에 따라 <strong>계획을 동적으로 수정</strong>합니다. 장기 목표(long-horizon task) 처리가 가능합니다.</p>
</div>
<div class="arch-diagram">
<div class="arch-planning">
<div class="planning-phase planning-plan">
<div class="planning-icon">P</div>
<h5>Plan</h5>
<p>작업 분해</p>
<div class="planning-tasks">
<span>Task 1</span>
<span>Task 2</span>
<span>Task 3</span>
<span>Task 4</span>
</div>
</div>
<div class="planning-arrows">
<div class="planning-arrow-down"></div>
</div>
<div class="planning-phase planning-execute">
<div class="planning-icon">E</div>
<h5>Execute</h5>
<p>단계별 ReAct 실행</p>
</div>
<div class="planning-arrows">
<div class="planning-arrow-down"></div>
</div>
<div class="planning-phase planning-reflect">
<div class="planning-icon">R</div>
<h5>Reflect</h5>
<p>결과 평가 & 계획 수정</p>
</div>
<div class="planning-replan">
<div class="replan-arrow"></div>
<span>Re-plan</span>
</div>
</div>
</div>
<div class="example-cards">
<div class="example-card example-card-wide">
<h5>"경쟁사 분석 보고서 만들어줘"</h5>
<div class="plan-execution">
<div class="plan-step plan-done">
<span class="plan-check">✓</span>
<span>1. 경쟁사 리스트 확정</span>
</div>
<div class="plan-step plan-failed">
<span class="plan-x">✗</span>
<span>2. 재무 데이터 수집</span>
<span class="plan-note">회사 B 데이터 못 찾음</span>
</div>
<div class="plan-step plan-replan">
<span class="plan-refresh">↻</span>
<span>2'. Re-plan: 대안 소스에서 재검색</span>
</div>
<div class="plan-step plan-done">
<span class="plan-check">✓</span>
<span>2'. 대안 소스에서 수집 완료</span>
</div>
<div class="plan-step plan-pending">
<span class="plan-dot">●</span>
<span>3. SWOT 분석</span>
</div>
<div class="plan-step plan-pending">
<span class="plan-dot">●</span>
<span>4. 보고서 작성</span>
</div>
</div>
</div>
</div>
</div>
</div>
<!-- Level 6 -->
<div class="level-card" id="level-6" data-level="6">
<div class="level-header">
<div class="level-badge">L6</div>
<div class="level-info">
<h3>Multi-Agent System</h3>
<div class="level-tags">
<span class="tag tag-pink">Collaboration</span>
<span class="tag tag-pink">Specialization</span>
<span class="tag tag-pink">Distributed</span>
</div>
</div>
<div class="level-autonomy">
<span class="autonomy-label">자율성</span>
<div class="autonomy-bar"><div class="autonomy-fill" style="width:95%"></div></div>
</div>
</div>
<div class="level-body">
<div class="level-desc">
<p>여러 에이전트가 각자의 역할, 도구, 프롬프트를 가지고 <strong>협업</strong>합니다. Orchestrator가 작업을 분배하고 결과를 통합하며, 각 에이전트가 독립된 컨텍스트를 가져 <strong>컨텍스트 윈도우 한계를 완화</strong>합니다.</p>
</div>
<div class="arch-diagram">
<div class="arch-multi-agent">
<div class="ma-orchestrator">
<div class="ma-node ma-orch">
<div class="ma-icon">O</div>
<span>Orchestrator</span>
</div>
</div>
<div class="ma-connections">
<div class="ma-line ma-line-1"></div>
<div class="ma-line ma-line-2"></div>
<div class="ma-line ma-line-3"></div>
<div class="ma-line ma-line-4"></div>
</div>
<div class="ma-workers">
<div class="ma-worker">
<div class="ma-node ma-research">
<div class="ma-icon">R</div>
<span>Researcher</span>
</div>
<div class="ma-tools">Search, Fetch</div>
</div>
<div class="ma-worker">
<div class="ma-node ma-coder">
<div class="ma-icon">C</div>
<span>Coder</span>
</div>
<div class="ma-tools">IDE, Terminal</div>
</div>
<div class="ma-worker">
<div class="ma-node ma-reviewer">
<div class="ma-icon">Q</div>
<span>Reviewer</span>
</div>
<div class="ma-tools">Lint, Test</div>
</div>
<div class="ma-worker">
<div class="ma-node ma-writer">
<div class="ma-icon">W</div>
<span>Writer</span>
</div>
<div class="ma-tools">Docs, Format</div>
</div>
</div>
</div>
</div>
<!-- Multi-agent patterns -->
<div class="patterns-grid">
<div class="pattern-card">
<h5>Orchestrator-Worker</h5>
<p>중앙 관리자가 작업을 분배하고 결과를 통합. 가장 일반적인 패턴.</p>
<div class="pattern-example">Claude Code의 sub-agents</div>
</div>
<div class="pattern-card">
<h5>Evaluator-Optimizer</h5>
<p>Generator가 생성, Evaluator가 평가 후 피드백. 반복적 개선.</p>
<div class="pattern-example">코드 리뷰 자동화</div>
</div>
<div class="pattern-card">
<h5>Debate / Adversarial</h5>
<p>Agent A가 주장, Agent B가 반론. Moderator가 최종 판정.</p>
<div class="pattern-example">의사결정 지원 시스템</div>
</div>
</div>
</div>
</div>
<!-- Level Comparison Table -->
<div class="comparison-table-section">
<h3 class="subsection-title">단계별 비교 요약</h3>
<div class="table-wrapper">
<table class="level-table">
<thead>
<tr>
<th>레벨</th>
<th>자율성</th>
<th>도구 사용</th>
<th>흐름 결정</th>
<th>대표 기술</th>
</tr>
</thead>
<tbody>
<tr>
<td><span class="table-badge">L0</span></td>
<td>없음</td>
<td>없음</td>
<td>코드</td>
<td>ChatGPT 기본 호출</td>
</tr>
<tr>
<td><span class="table-badge">L1</span></td>
<td>낮음</td>
<td>단일 턴</td>
<td>코드</td>
<td>RAG, Function Calling</td>
</tr>
<tr>
<td><span class="table-badge">L2</span></td>
<td>낮음</td>
<td>순차</td>
<td>코드</td>
<td>LangChain Chains</td>
</tr>
<tr>
<td><span class="table-badge">L3</span></td>
<td>중간</td>
<td>분기</td>
<td>코드+LLM</td>
<td>Semantic Router</td>
</tr>
<tr class="table-highlight">
<td><span class="table-badge badge-highlight">L4</span></td>
<td>높음</td>
<td>루프</td>
<td>LLM</td>
<td>ReAct, Claude Tool Use</td>
</tr>
<tr>
<td><span class="table-badge">L5</span></td>
<td>높음</td>
<td>계획+루프</td>
<td>LLM</td>
<td>Plan-and-Execute, ADK</td>
</tr>
<tr>
<td><span class="table-badge">L6</span></td>
<td>매우 높음</td>
<td>분산</td>
<td>다수 LLM</td>
<td>CrewAI, AutoGen</td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</section>
<!-- Architecture Deep Dive -->
<section class="section" id="architecture">
<div class="container">
<div class="section-header">
<span class="section-label">Deep Dive</span>
<h2 class="section-title">핵심 아키텍처 개념</h2>
<p class="section-desc">에이전트 시스템의 내부를 구성하는 핵심 개념들</p>
</div>
<!-- Memory Architecture -->
<div class="deep-dive-card">
<div class="dd-header">
<h3>Agent Memory 시스템</h3>
<span class="dd-tag">Core Component</span>
</div>
<div class="dd-body">
<p>인간의 기억 체계에서 영감을 받은 에이전트 메모리 시스템은 세 가지 유형으로 나뉩니다.</p>
<div class="memory-arch">
<div class="memory-type">
<div class="memory-icon memory-short">S</div>
<h4>Short-term Memory</h4>
<p>컨텍스트 윈도우 내의 작업 기억. 현재 대화와 즉시 필요한 정보를 유지합니다.</p>
<div class="memory-impl">구현: Context Window, Working Memory</div>
</div>
<div class="memory-type">
<div class="memory-icon memory-long">L</div>
<h4>Long-term Memory</h4>
<p>세션을 넘어 지속되는 기억. 사실, 정의, 규칙 등 구조화된 지식을 저장합니다.</p>
<div class="memory-impl">구현: Vector DB, Knowledge Graph</div>
</div>
<div class="memory-type">
<div class="memory-icon memory-episodic">E</div>
<h4>Episodic Memory</h4>
<p>과거 경험과 에피소드를 기록. 유사한 상황에서 과거 경험을 참조합니다.</p>
<div class="memory-impl">구현: Vector DB + Semantic Retrieval</div>
</div>
</div>
<div class="dd-source">Source: Park et al., "Generative Agents" (2023) | IBM, "AI Agent Memory" (2025)</div>
</div>
</div>
<!-- Agentic RAG -->
<div class="deep-dive-card">
<div class="dd-header">
<h3>Agentic RAG vs Traditional RAG</h3>
<span class="dd-tag">Evolution</span>
</div>
<div class="dd-body">
<div class="rag-comparison">
<div class="rag-type rag-traditional">
<h4>Traditional RAG</h4>
<div class="rag-flow">
<div class="rag-step">Query</div>
<div class="rag-arrow">↓</div>
<div class="rag-step">Vector Search</div>
<div class="rag-arrow">↓</div>
<div class="rag-step">Retrieve Docs</div>
<div class="rag-arrow">↓</div>
<div class="rag-step">Generate Answer</div>
</div>
<p class="rag-desc">단일 패스. 검색 결과가 부족해도 재시도 없음. 도서관에서 책 한 권 빌리는 것과 같음.</p>
</div>