AI product & agentic workflows
Turning model capability into bounded product behavior, review paths, and reusable operating patterns.
Career-facing view
Wonbeen Lee
AI / Global Product Manager
I plan and ship products in complex B2B SaaS environments, with a growing focus on Agentic AI. I am strongest at turning ambiguous product problems into clear product direction, requirements, workflows, and repeatable execution systems.
Professional focus
What I work on.
The work I am strongest at spans four connected areas.
Enterprise SRM / S2P SaaS
Product planning experience in complex B2B procurement contexts where workflows, permissions, and cross-functional execution matter.
Product Ops & execution systems
Structuring ambiguous work into clearer intake, scope, evidence, review, and release loops rather than relying on one-off heroics.
AI-ready knowledge systems
Designing knowledge so people and agents can find, interpret, reuse, and safely publish the same public source.
Career snapshot
Global product planning in enterprise procurement SaaS.
At emro, I work on global B2B SaaS product planning across SRM / S2P contexts. My scope spans turning incoming needs into product direction and requirements, coordinating cross-functional delivery, and increasingly shaping Agentic AI product experiences.
Responsibility signals
- Product planningFrame ambiguous needs into scope, priorities, and decision-ready requirements.
- Cross-functional deliveryWork across product, design, engineering, and QA from planning through release.
- AI product translationTranslate model capability into bounded user flows, review paths, and operating workflows.
Customer details, internal roadmaps, private metrics, and employer-confidential context stay outside this public portfolio.
Read the public-work boundary →Selected work
A few examples of how I turn ambiguous product problems into working systems.
Three projects across AI workflows, product operations, and AI-ready knowledge systems — showing the context, my role, what I built, and the public evidence behind it.
Knowledge product · Product system
Personal AI Operating Site
- Context
- Professional knowledge, experiments, and reusable AI work methods become hard to trust when they are scattered across tools and conversations.
- My role
- Creator / product owner
- What I did
- Designed and operate a public-safe, typed knowledge product that connects editorial content, machine-readable indexes, Workflows, Skills, and a Knowledge Map.
- Outcome / signal
- End-to-end product ownership across product framing, information architecture, AI-readability, publishing boundaries, and ongoing operation.
AI workflow product · Builder
Prompt Hotbar
- Context
- AI-heavy work creates a small but persistent retrieval problem: the same approval, scope-control, verification, and handoff instructions get retyped or hunted down repeatedly.
- My role
- Creator / product owner
- What I did
- Turned that friction into a keyboard-first prompt-reuse product, then expanded it through real use into reviewed discovery surfaces, Workflow Sets, browser-local personalization, and a bounded local MCP pilot.
- Outcome / signal
- Product discovery through use: repeated workflow friction became a focused tool, then expanded only where recurring behavior justified a broader retrieval system.
Agent workflow · Open-source tool
Universal Context Handoff
- Context
- AI coding sessions can preserve code while losing the decisions, constraints, rejected directions, and next actions needed by the next agent or session.
- My role
- Creator / workflow designer
- What I did
- Built a local-first CLI and assistant-skill experiment that converts session context into structured Markdown and JSON handoff artifacts with explicit redaction and human review boundaries.
- Outcome / signal
- Workflow design under AI constraints: context loss becomes a portable, reviewable artifact instead of a hidden dependency on one session or agent.
How I work
The operating system is part of the proof.
PAOS makes some of my working principles inspectable instead of describing them only as traits.
Start from the state, not the tool.
The Workflow Router begins with the work state — Clarify, Plan, Execute, Verify, Publish & Operate — before recommending an asset.
See Workflow Router →Make verification observable.
Evidence-first Review treats a plausible completion claim as a hypothesis until the output, checks, and boundaries can be inspected.
See Evidence-first Review →Keep public proof bounded.
Public-safe Publishing makes the privacy boundary part of the workflow instead of relying on cleanup after the fact.
See Public-safe Publishing →Resume
Current public career record.
A current downloadable resume is not published here yet. LinkedIn is the current public career record while I prepare a canonical resume for this surface.
View LinkedIn profileOpportunity fit
Where the next conversation is most relevant.
I am most interested in product roles and collaborations where AI-native workflows, complex B2B SaaS, and global product execution overlap — especially where product judgment matters as much as model capability.
Connect on LinkedIn Collaboration scope
Recruiting/career context lives here; broader writing, speaking, coffee-chat, and workflow conversations stay in Collaboration.