Career
AI / Global Product Manager
Resume, detailed experience, and selected professional cases in one career-facing view.
- 01 Career
- 02 Resume / CV
- 03 Career history
- 04 Cases
A personal knowledge product for AI‑era product work.
Essays, copy-ready prompts and templates, lab notes, and machine-readable indexes from one product manager's AI operating system — public notes on AI Product Ops, agentic workflows, and AI-ready knowledge systems.
Articles · Prompts · Templates · Labs · Projects · Skills · llms.txt Who’s behind this?Agent Society
A small society behind the work.
I don’t use one assistant for everything. Personal and Work agents divide research, routing, memory, watching, publishing, and execution — each with a durable role and a small motion identity.
PERSONAL 10 WORK 4
Meet the agents → Featured knowledge streams
Recent work across six content types.
Everything on this site is typed — article, prompt, template, lab note, project, or reference — so both readers and agents know what they are looking at.
From Prompt Reuse to Personal Retrieval: Preference Memory Without Accounts
Finding a useful item once did not make it easier to return to the items that mattered most, but solving that problem with accounts or a backend profile would add complexity before it was needed.
Read article →Agentic Handoff Brief
Package the current state of a task so another agent (or a future you) can continue without re-deriving context.
- Input
- task context, current progress
- Output
- structured markdown brief
AI-Readable Site IA Planner
- Sections
- 7 sections
- Use case
- Plan a site's human-readable and machine-readable structure together: routes, content types, llms.txt, and markdown mirrors.
- Related prompt
- agentic handoff brief
From Canonical Profile to Essence: Designing a Motion-Ready Identity System for AI Agents
- Hypothesis
- A durable agent identity can survive in a small UI surface if silhouette, color ownership, signature motif, and motion grammar are treated as one identity system rather than as decoration around a portrait.
- Tool / setup
- Astro · CSS · SVG · vanilla JavaScript — Fourteen public-safe Agent Society identities were distilled from Canonical Profiles into small Essence forms, then implemented with autonomous idle behavior, local desktop cursor awareness, mobile autonomous motion, and reduced-motion fallbacks.
- Result state
- observed
- Next iteration
- Test the same Essence identities on lower-density surfaces such as Home and operating navigation without letting motion compete with the site's knowledge content.
- Artifact
- Interactive artifact →
Prompt Hotbar
- Status
- active
- What it tests
- A keyboard-first workspace for retrieving reusable AI instructions and reviewed operational references at the moment they are needed.
- Public artifact
- tool
- Related
- a prompt library is not enough · ai product planning team capability · evidence first ai workflow · prompt hotbar local mcp pilot · prompt hotbar local first preference layer
The llms.txt Pattern
- Source
- website — llmstxt.org
- Why it matters
- It reframes AI-readability as curation: a site tells agents what to read instead of hoping crawlers guess well.
Library
Reusable assets for the work in front of you.
Each library item ships with its use case, required input, and expected output — ready to copy into real work.
Prompts
/promptsTemplates
/templatesReferences
/referencesThis library starts small. Each item is meant to be reusable, public-safe, and can be connected to a real workflow when the next action calls for it.
Lab notebook
Experiments, recorded like experiments.
Every lab note keeps the same structure — hypothesis, setup, result, next iteration — so the record stays honest and comparable.
Chat Conversation Mining for Public Knowledge
- Hypothesis
- Long AI chat sessions contain publishable assets worth extracting.
- Tool / setup
- Conversation-to-Source Extractor — Extraction prompt run over exported working sessions, outputs staged for boundary review
- Result state
- observed
- Next iteration
- Automate the public/private boundary check before publishing instead of reviewing manually.
From Canonical Profile to Essence: Designing a Motion-Ready Identity System for AI Agents
- Hypothesis
- A durable agent identity can survive in a small UI surface if silhouette, color ownership, signature motif, and motion grammar are treated as one identity system rather than as decoration around a portrait.
- Tool / setup
- Astro · CSS · SVG · vanilla JavaScript — Fourteen public-safe Agent Society identities were distilled from Canonical Profiles into small Essence forms, then implemented with autonomous idle behavior, local desktop cursor awareness, mobile autonomous motion, and reduced-motion fallbacks.
- Result state
- observed
- Next iteration
- Test the same Essence identities on lower-density surfaces such as Home and operating navigation without letting motion compete with the site's knowledge content.
- Artifact
- Interactive artifact →
Multiplatform Context Handoff
- Hypothesis
- AI coding session의 결과물뿐 아니라 작업 맥락, 금지사항, 검증 결과, 다음 행동을 구조화해 남기면 다른 agent나 repo로 작업을 더 안전하게 이어갈 수 있다.
- Tool / setup
- ChatGPT · Claude Code · Codex CLI · Antigravity · Markdown · JSONL · GitHub — ChatGPT(전략/prompt), Claude Code(구현), Codex CLI(생성/검증), Antigravity(IDE 실행) 등 멀티 플랫폼 환경 간 handoff artifacts 전달 실험
- Result state
- observed
- Next iteration
- Agentic Handoff Brief template과 Universal Context Handoff CLI를 함께 사용해 repo-to-repo handoff 품질을 비교한다.
Personal AI Operating Site 1-day Prototype
- Hypothesis
- A personal site can read as a knowledge product, not a portfolio.
- Tool / setup
- Fable 5 · Astro — single-page prototype → external evaluation → DESIGN.md → Astro scaffold
- Result state
- running
- Next iteration
- Compare the v0.2 scaffold against acceptance criteria; fix whatever still feels portfolio-ish.
Public Knowledge Router: Turning Raw AI Conversations into Publishable Knowledge Assets
- Hypothesis
- The public/private routing pipeline can be shown, not just described, in a static-site interactive.
- Tool / setup
- Astro · vanilla JS — Self-contained Astro component at /labs/public-knowledge-router — keyboard-first routing, pointer drag as enhancement, agent-view pipeline, SVG asset map; no backend.
- Result state
- running
- Next iteration
- Collect feedback on whether the agent view makes the markdown-mirror and llms.txt concepts legible to first-time visitors.
- Artifact
- Interactive artifact →
Projects
Longer arcs the lab notes feed into.
Projects here are running experiments with public artifacts — not finished case studies.
Personal Agent Society — A bounded multi-agent operating model
- Status
- active
- What it tests
- Why I split AI collaboration across durable roles, environments, relationships, and authority boundaries instead of treating one generic assistant as the whole system.
- Public artifact
- website
- Related
- from canonical profile to essence · designing agents with boundaries not just personas
Personal AI Operating Site
- Status
- active
- What it tests
- A public-safe personal knowledge product that serves human readers and AI agents from the same typed content source.
- Public artifact
- website
- Related
- personal ai operating site one day prototype · why im building a personal ai operating site
Universal Context Handoff
- Status
- active
- What it tests
- A local-first CLI and assistant skill for preserving AI coding session context across repositories and tools like Claude Code, Codex CLI, and Antigravity. Built as a practical handoff protocol, not a cloud memory system.
- Public artifact
- tool
- Related
- agentic handoff brief
Prompt Hotbar
- Status
- active
- What it tests
- A keyboard-first workspace for retrieving reusable AI instructions and reviewed operational references at the moment they are needed.
- Public artifact
- tool
- Related
- a prompt library is not enough · ai product planning team capability · evidence first ai workflow · prompt hotbar local mcp pilot · prompt hotbar local first preference layer
More project objects land here as experiments mature. Each one arrives with a status, a public artifact, and the lab notes that fed it.
Publishing boundary
Every note passes a boundary check before it ships.
This is the public layer of a private knowledge system. Each item is classified before publishing — that is what keeps this layer safe to read, copy, and crawl.
public
Publishable as-is. Ships to pages, feeds, and the AI-readable layer.
generalized
Publishable after company and personal context is removed and generalized.
review
Needs a human boundary check before anything ships.
private-only
Never published. Stays in the private knowledge base, full stop.
AI-readable layer
Two layers, one source of truth.
This site has a human-readable layer and a machine-readable layer. AI
agents should start from /llms.txt and use only public,
curated content.
agent read path /llms.txt /llms-full.txt /content-graph.json /md/[type]/[slug].md canonical page
Public, curated content only. The private knowledge base never ships to this layer.
/llms.txt
# Personal AI Operating Site > Public-facing, AI-readable knowledge product focused on > AI Product Ops, agentic workflows, reusable prompts/templates, > and public lab notes. ## Start here - /articles - /notes - /prompts - /templates - /labs - /projects ## Policies - Public, curated content only - Markdown mirror available under /md/ - Full index: /llms-full.txt ## Boundary - visibility: public | generalized only - private-source and private-only never ship
Collaboration
Open to good conversations.
If the notes here overlap with problems you are working on, a short conversation is welcome. Read the fit list first — it saves both of us time.
Good fit
- AI Product Ops and execution systems
- Agentic workflows and MCP in real product work
- AI-ready knowledge systems for teams
- Practical AI adoption for product teams
Not a fit
- Generic AI hype content or ghost-written thought leadership
- Building login/payment/community platforms
- Anything that requires sharing private or client-confidential context