2nd/10_Wiki/Topics/AI_and_ML/Collaborative Programming (Pair & Mob).md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-collaborative-programming	Collaborative Programming (Pair & Mob)	10_Wiki/Topics	verified	self	pair programming mob programming ensemble programming driver-navigator real-time collaboration	none	A	0.88	applied	collaborative-programming pair-programming mob-programming ensemble agile knowledge-transfer code-review		2026-05-10	pending	language framework process Live Share / Tuple / Pop / Replit Multiplayer

Collaborative Programming (Pair & Mob)

매 한 줄

"매 real-time review + 매 collective knowledge". 매 PR 의 lag 의 X — 매 instant. 매 pair (2) / mob (3+, ensemble). 매 modern: 매 LLM-aided pair (Cursor pair with AI). 매 AI 시대 의 still relevant — 매 design decision 의 human collective.

매 핵심

Pair Programming

Driver-Navigator

• Driver: 매 keyboard, 매 tactical.
• Navigator: 매 strategic, 매 review.
• Switch: 매 5-25 분.

Ping-pong (TDD-paired)

• A 매 test, B 매 implement.
• 매 switch.

Strong-style

• "For an idea to go from your head into the computer, it MUST go through someone else's hands."
• → 매 navigator 의 thinking, 매 driver 의 typing 만.

Mob (Ensemble) Programming

• 매 3+ developer + 매 1 keyboard.
• 매 매 short rotation (4-10 분).
• 매 모든 brain 의 active.
• 매 high-stakes / complex problem.

매 benefit

1. Real-time review: 매 0-lag.
2. Knowledge transfer: 매 senior 의 tacit.
3. Reduced bus factor.
4. Onboarding: 매 fast.
5. Decision quality.
6. Less context switch (one task focused).

매 cost

1. Resource: 매 N person 의 1 task.
2. Fatigue: 매 high cognitive load.
3. Personality fit.
4. Remote 의 어려움 (overcomeable).
5. Boring task 의 over-engineer 의 risk.

매 effective use case

Pair

• Complex bug.
• New feature design.
• Onboarding (junior + senior).
• Cross-functional (frontend + backend).
• Refactoring critical area.
• Security-sensitive code.

Mob

• Architecture decision.
• Domain modeling.
• New dev environment setup.
• Big incident response.
• Knowledge crystallization.

매 not effective

• 매 simple CRUD.
• 매 documentation.
• 매 mechanical refactor.
• 매 deep flow individual work.

매 modern tool

• VS Code Live Share: 매 Microsoft.
• Tuple: 매 macOS pair.
• Pop: 매 web-based.
• Replit Multiplayer: 매 cloud IDE.
• JetBrains Code With Me: 매 IntelliJ family.

매 LLM-aided pair (2024+)

• 매 Cursor: 매 AI 의 pair.
• 매 Copilot: 매 AI 의 navigator hint.
• 매 Cline / Aider: 매 agentic.

→ 매 human-AI pair 의 새 form.

매 timebox / cadence

Pair

• 매 25 min on / 5 min break (Pomodoro).
• 매 4-6 hour / day max.
• 매 2 person 의 stable cadence.

Mob

• 매 5-10 min driver rotation.
• 매 60 min session + 10 min break.
• 매 1 day max with breaks.

💻 패턴 (응용)

Pair session (Live Share)

# 매 host
code .  # 매 VS Code
# 매 Live Share extension
# 매 "Start Collaboration Session"
# 매 share link

# 매 guest
# 매 join via link
# 매 follow / lead 의 toggle

Mob rotation timer (web)

// 매 mob rotation timer
class MobTimer {
  constructor(participants, intervalMin = 5) {
    this.participants = participants;
    this.interval = intervalMin * 60 * 1000;
    this.currentDriver = 0;
  }
  
  start() {
    this.timerId = setInterval(() => this.rotate(), this.interval);
    this.notifyDriver(this.participants[this.currentDriver]);
  }
  
  rotate() {
    this.currentDriver = (this.currentDriver + 1) % this.participants.length;
    this.notifyDriver(this.participants[this.currentDriver]);
  }
  
  notifyDriver(name) {
    notify(`${name} is now driving for ${this.interval / 60000} min`);
  }
}

Strong-style discipline

Rule: "For an idea to go from your head into the computer,
       it MUST go through someone else's hands."

Driver: 매 type 만 — 매 idea 의 X.
Navigator: 매 think 만 — 매 keyboard 의 X.

Switch driver 매 25 min.

Ping-pong TDD

Person A:
  - Write a failing test.
  - Pass keyboard.

Person B:
  - Write minimum code to pass.
  - Refactor.
  - Write next failing test.
  - Pass keyboard.

Person A:
  - Pass test, refactor, write next test.
  - ...

LLM-aided pair (Cursor)

Human (driver / navigator):
  - High-level intent.
  - Architecture decision.
  - Final review.

AI (Cursor):
  - Code generation suggestion.
  - Refactor proposal.
  - Test scaffolding.
  - Documentation draft.

Human-AI handoff every few cycles.
Human always commits — AI doesn't.

Session retro (10 min)

session_retro:
  - what_worked: '...'
  - what_didnt: '...'
  - knowledge_transferred:
    - 'X learned about Y'
    - 'Z explained the W pattern'
  - decisions_made:
    - 'Use library X over Y because Z'
  - next_steps:
    - '...'

Pair / mob effectiveness measurement

def measure_collab_value(pair_sessions, control_sessions):
    """매 RCT-light comparison."""
    return {
        'bug_density_pair': bug_count(pair_sessions) / loc(pair_sessions),
        'bug_density_solo': bug_count(control_sessions) / loc(control_sessions),
        'review_time_pair': median([s.review_time for s in pair_sessions]),
        'review_time_solo': median([s.review_time for s in control_sessions]),
        'lead_time_pair': median([s.lead_time for s in pair_sessions]),
        'lead_time_solo': median([s.lead_time for s in control_sessions]),
        'knowledge_transfer_score': survey([s.team for s in pair_sessions]),
    }

Hybrid policy (decision tree)

def should_pair(task):
    if task.complexity >= 'high': return True
    if task.security_sensitive: return True
    if task.crosses_module_boundary: return True
    if task.assigned_to_junior and team.has_senior_available(): return True
    if task.architecture_decision: return 'mob'  # 매 mob 의 better
    return False  # 매 solo + async review

🤔 결정 기준

상황	Mode
Complex bug	Pair
Architecture	Mob
Onboarding	Pair (senior + junior)
Routine CRUD	Solo + async review
Documentation	Solo
Cross-functional	Pair
Incident response	Mob
Mechanical refactor	Solo

기본값: 매 hybrid — 매 pair / mob 의 high-stakes, 매 async 의 routine.

🔗 Graph

• 부모: Agile
• 변형: Pair-Programming · Mob-Programming
• 응용: Cursor (AI pair)
• Adjacent: Quality_Code_Review_Modern · Asset-Specific-Knowledge · Cognitive Constraints · Codebase_Onboarding_Guide

🤖 LLM 활용

언제: 매 team practice. 매 onboarding. 매 critical work. 매 AI pair design. 언제 X: 매 routine work (overhead). 매 introvert-only team (force X).

❌ 안티패턴

• 모든 task 의 pair: 매 cost.
• No rotation in mob: 매 1 driver 만.
• No timebox: 매 burnout.
• No retro: 매 learning lose.
• Pair 의 unequal participation (silent navigator): 매 cost X.
• AI pair 의 sole reliance: 매 human collective lose.

🧪 검증 / 중복

• Verified (Williams pair programming research, Zuill mob programming, GitHub Engineering case studies).
• 신뢰도 A.
• Related: Quality_Code_Review_Modern · Codebase_Onboarding_Guide · Cognitive Constraints · Asset-Specific-Knowledge · Branching Strategies.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — pair / mob style + LLM pair + 매 timer / TDD / measurement code