Wonbeen Lee wbeen / AI Product Ops notebook

Lab notebook

Experiments, recorded like experiments.

Public lab notes keep the same operating shape — hypothesis, setup, result, and next iteration — so experiments stay comparable and reusable instead of turning into polished hindsight.

5 public lab notes currently available.

Lab 01

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 →
View lab note →
2026-09-03 #agent-ux#visual-identity
Lab 02

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 →
View lab note →
2026-07-08 #knowledge-capture#publishing
Lab 03

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.
View lab note →
2026-07-07 #agentic-workflows#context-handoff
Lab 04

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.
View lab note →
2026-07-04 #knowledge-capture#extraction
Lab 05

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.
View lab note →
2026-07-04 #prototype#knowledge-product