기능적으로 정말 차별하고 싶어.
Captured from ChatGPT on 5/22/2026
User
기능적으로 정말 차별하고 싶어.
ChatGPT
기능 차별화는 “기능 개수”가 아니라 AI가 실제로 더 잘 작동하는가로 결정돼.
지금 시장의 대부분은:
-
노트 앱
-
RAG wrapper
-
vector DB UI
-
graph visualization
-
MCP connector catalog
이 중 하나야.
근데 memory.wiki가 진짜 커지려면 아래 중 최소 하나를 독점적으로 잘해야 해:
1. “AI-optimized context packaging”
이게 가장 중요하다.
지금 세상은 문서가 인간용으로 작성돼 있음.
근데 LLM은:
-
context window 제한 있음
-
relevance ranking imperfect
-
stale context 문제 있음
-
narrative continuity 약함
-
intent-aware compression 못함
즉 AI는 “raw docs”를 잘 못 먹는다.
그래서 memory.wiki의 핵심은:
저장이 아니라 “AI consumption optimization”
이어야 함.
여기서 진짜 차별 기능이 나온다
A. Intent-Adaptive Bundles
이건 매우 강력해질 수 있음.
사용자가 같은 graph를 갖고 있어도:
-
투자자 미팅용
-
코드 작업용
-
논문 작성용
-
법률 검토용
-
onboarding용
context가 달라야 함.
그러니까:
memory.wiki/project-x?intent=coding
memory.wiki/project-x?intent=fundraising
memory.wiki/project-x?intent=research
를 만들면,
AI가 읽는 bundle이 달라짐.
이건 단순 retrieval이 아님.
이건:
-
context hierarchy
-
relevance shaping
-
narrative ordering
-
token budgeting
-
contradiction filtering
을 포함함.
여기서부터 “AI OS” 느낌이 난다.
2. Temporal Memory
이건 거의 아무도 제대로 안 함.
LLM context의 가장 큰 문제 중 하나:
시간이 없다.
AI는:
-
무엇이 최신인지
-
어떤 결정이 deprecated됐는지
-
왜 바뀌었는지
를 잘 모른다.
memory.wiki는 graph edge에:
-
valid_from
-
valid_to
-
supersedes
-
confidence
-
source
를 넣을 수 있음.
그러면:
“현재 기준으로만 설명해줘” “지난달 기준 전략 보여줘” “왜 architecture 바뀌었는지 설명해줘”
가 가능해짐.
이건 단순 note app이 절대 못 한다.
3. Contradiction-aware memory
이거 진짜 중요하다.
대부분 RAG는:
-
모든 걸 retrieval함
-
conflicting info 같이 줌
-
hallucination 증가
근데 실제 인간 memory는:
-
old belief
-
discarded idea
-
current truth
를 구분한다.
memory.wiki는:
-
archived beliefs
-
rejected decisions
-
experimental hypotheses
를 구조적으로 저장 가능해야 함.
예:
This architecture was abandoned because:
- scaling issue
- auth complexity
- token cost
AI가 reasoning quality가 확 올라간다.
4. AI-facing representation layer
이건 진짜 중요.
인간용 markdown ≠ AI optimal format
memory.wiki는 내부적으로:
-
graph
-
entity map
-
semantic edges
-
episodic logs
-
summaries
-
canonical definitions
를 저장하고,
AI에겐 dynamic compiled view를 보여줘야 함.
즉:
Human View
AI View
Condensed View
Chronological View
Decision View
가 다 달라야 함.
이건 그냥 wiki가 아니다.
5. “Canonical Self”
이건 매우 강력한 방향 가능성 있음.
지금 AI의 큰 문제:
매 세션마다 사용자가 다시 자신을 설명함.
memory.wiki는:
/about /current-focus /preferences /writing-style /project-state /decision-history
같은 canonical identity layer를 만들 수 있음.
그리고 모든 AI가 이것을 참조.
이건:
-
persistent persona
-
working memory
-
epistemic continuity
를 제공한다.
엄청 큰 시장이다.
6. Agent-ready Memory
진짜 장기적으로 가장 중요.
미래 AI는:
-
단일 chat app이 아니라
-
여러 specialized agents ecosystem
으로 갈 가능성이 높다.
그러면 필요한 건:
shared memory substrate
다.
memory.wiki가:
-
agent-readable
-
structured
-
permissioned
-
streamable
-
update-aware
memory layer가 되면,
AI ecosystem infra가 된다.
여기서 진짜 커진다.
7. 너희가 반드시 피해야 할 함정
“그래프 시각화”
이거 founder들은 좋아하는데,
대부분 사용자 가치 거의 없음.
graph visualization은:
-
retention driver 아님
-
PMF driver 아님
-
workflow core 아님
이다.
사람은 “보는 graph”보다:
-
AI 결과 품질 향상
-
context loss 감소
-
repeated explanation 제거
를 원함.
8. 가장 중요한 differentiation
결국 핵심은 이것 하나다:
“memory.wiki URL을 넣으면 AI output quality가 실제로 좋아진다.”
이게 observable해야 함.
그러니까 앱의 중심 metric은:
-
uploads 수
-
nodes 수
-
graph density
가 아니라,
“Context Success Rate”
같은 게 돼야 함.
예:
-
repeated clarification 감소율
-
AI task completion 증가
-
hallucination 감소
-
onboarding latency 감소
이런 것.
내가 생각하는 가장 강한 positioning
지금:
AI knowledge graph
아님.
대신:
AI-native memory infrastructure
또는
Context operating system for AI
또는
Shared memory layer across AI tools
이 더 강하다.
그리고 마지막으로 제일 중요:
절대 “AI second brain”으로 브랜딩하지 마.
너무 crowded다.
대신:
“The memory format AI actually understands.”
이 훨씬 새롭다.