Files

T

Antigravity Agent 504fd5fb42 [G1-Sync] Manual knowledge update

2026-05-10 22:08:15 +09:00

6.5 KiB

Raw Blame History

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

title

Strategic Thinking

매 한 줄

"매 long-horizon, multi-step, 매 trade-off 의 reasoning — 매 goal → diagnosis → guiding policy → coherent actions". Richard Rumelt, Good Strategy / Bad Strategy (2011) 의 kernel. 매 engineering context — 매 architecture choice, tech debt, hiring, platform investment.

매 핵심

매 Rumelt kernel

Diagnosis: 매 challenge 의 명확한 정의 — what is actually hard?
Guiding Policy: 매 approach — how we'll address it.
Coherent Actions: 매 mutually reinforcing concrete steps.

매 strategic vs tactical

Strategic: months-years, irreversible, enables/blocks options.
Tactical: days-weeks, reversible, executes within strategy.
Type 1 vs Type 2 decisions (Bezos): 매 one-way door (strategic) vs two-way door (tactical).

매 frameworks

OKRs: Objective + Key Results.
Wardley Map: 매 capability evolution chain (genesis → custom → product → commodity).
Five Forces (Porter): 매 industry structure.
First Principles (Musk): 매 axiom 부터 reason.
Pre-mortem (Klein): 매 future failure 의 imagine.

매 응용

Engineering org strategy (build vs buy, monolith vs micro).
Career planning (5-year arc).
Technical roadmap prioritization.

💻 패턴

Pre-mortem template (Markdown ADR)

# ADR-042: Migrate to Kubernetes

## Diagnosis
Current ECS deployment cannot scale to multi-region; deploys take 40min.

## Pre-mortem (it's 2027, this failed)
- Team didn't learn k8s; constant outages.
- Cost 3× higher than ECS.
- Migration took 18 months instead of 6.

## Mitigations
- Hire 1 SRE with k8s background before start.
- Pilot with 2 non-critical services for 3 months.
- Define rollback criteria.

Wardley Map (text DSL — onlinewardleymaps.com)

title Engineering Platform 2026

component User [0.95, 0.50] label [0, -10]
component Web App [0.85, 0.55]
component Auth [0.75, 0.70]
component LLM Inference [0.55, 0.30] label [0, 10]
component GPU Cluster [0.30, 0.40]
component Datacenter [0.10, 0.30]

User -> Web App
Web App -> Auth
Web App -> LLM Inference
LLM Inference -> GPU Cluster
GPU Cluster -> Datacenter

First-principles decision (LLM-augmented)

# Use Claude Opus 4.7 for option exploration
from anthropic import Anthropic
client = Anthropic()
def explore_options(problem: str, constraints: list[str]) -> str:
    msg = client.messages.create(
        model="claude-opus-4-7",
        max_tokens=2000,
        thinking={"type": "enabled", "budget_tokens": 8000},
        messages=[{"role": "user", "content":
            f"Problem: {problem}\nConstraints: {constraints}\n"
            "From first principles, list 5 distinct approaches with key trade-offs."}],
    )
    return msg.content[-1].text

Decision matrix (weighted)

options = {
    "Postgres":  {"perf": 7, "cost": 9, "team_skill": 9, "scale": 6},
    "DynamoDB":  {"perf": 9, "cost": 5, "team_skill": 4, "scale": 9},
    "CockroachDB":{"perf": 8, "cost": 6, "team_skill": 5, "scale": 8},
}
weights = {"perf": 0.3, "cost": 0.2, "team_skill": 0.3, "scale": 0.2}
scores = {k: sum(v[c]*weights[c] for c in weights) for k, v in options.items()}
print(sorted(scores.items(), key=lambda x: -x[1]))

OKR structure

objective: Make platform self-serve for new teams
key_results:
  - 90% of new services deploy without platform team involvement (currently 20%)
  - Onboarding time < 1 day (currently 3 weeks)
  - NPS from internal devs > 50

Type-1 vs Type-2 framing

| Decision | Type | Reversibility | Rigor needed |
|---|---|---|---|
| Choose primary DB | 1 | Low | High — involve all leads |
| Choose CI provider | 2 | Medium | Medium |
| Choose linter ruleset | 2 | High | Low — just decide |

Strategy doc skeleton (Will Larson style)

# Strategy: Platform Reliability 2026

## Context
Outages weekly, MTTR 4hrs, eng team rotates on-call.

## Diagnosis
Lack of observability + flaky integration tests + no SLO discipline.

## Guiding Policy
Invest in observability and SLO culture before adding features.

## Actions
1. Adopt OpenTelemetry across all services (Q1).
2. Define SLOs for top 10 services (Q1).
3. Error-budget-based release gating (Q2).
4. Hire 2 SREs (Q2).

## Non-goals
- Multi-region active-active (defer to 2027).
- New feature work pause: NO — 70/30 split.

매 결정 기준

상황	Approach
Long-horizon org/tech direction	Rumelt kernel + Wardley map
Capability evolution planning	Wardley map
Multi-option comparison	Weighted decision matrix + ADR
Risk-laden irreversible decision	Pre-mortem + Type-1 framing
Quarterly execution	OKRs
Novel/uncertain space	First principles + LLM exploration

기본값: 매 strategic decision → ADR + diagnosis + pre-mortem.

🔗 Graph

부모: Decision Making · Engineering Leadership
변형: Wardley Mapping · OKRs · Pre-mortem
응용: Architecture Decision Record · Tech Strategy · Roadmap Planning
Adjacent: First Principles Thinking · Systems Thinking · Mental Models

🤖 LLM 활용

언제: option exploration, pre-mortem brainstorming, devil's advocate, summarize long context for diagnosis. 언제 X: 매 final commit decision — accountability 의 human; 매 confidential strategy — privacy concern.

❌ 안티패턴

Goals as strategy: "be #1" 의 strategy X — 매 goal. 매 diagnosis + policy 의 missing.
Strategy as wishlist: 매 incoherent action list, 매 trade-offs 의 unstated.
Pre-mortem 의 skip: 매 surprise failure mode.
Analysis paralysis on Type-2 decisions: 매 reversible 의 just decide.
Wardley map as drawing exercise: 매 decision 으로 connect X — 매 useless.

🧪 검증 / 중복

Verified (Rumelt Good Strategy/Bad Strategy 2011; Larson An Elegant Puzzle; Wardley Wardley Maps 2018).
신뢰도 B (synthesis from multiple authorities).

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — strategic thinking frameworks for engineering

6.5 KiB Raw Blame History Unescape Escape