What needs to stay human?
In automation work, trust starts with the boundary: what the system can do, what it must explain, and where a person can take control again.
Product systems · Design leadership · Human-centered AI · Tokyo
I design the systems around hard decisions.
The hiring flow that asks too much. The mobile product with too many signals. The team that needs a better way to think together. The AI agent that should not act without a human in charge.
The strongest evidence points to that kind of work: messy systems, consequential decisions, and problems a cleaner screen alone will not fix.
Belief
Technology should not make people smaller.
A product can be faster and still leave people unsure. A model can be powerful and still be impossible to trust. A dashboard can be full of evidence and still not help anyone decide.
I work around the screen as much as on it: what someone needs to understand, what the system is allowed to do, where control stays human, and what happens when the system gets it wrong.
Professional proof
This did not start with AI.
Indeed: employer hiring systems, recruiting automation, mobile strategy, research synthesis, design sprints, and product-learning programs.
SEEK Asia: regional marketplace product/design systems across self-service employer tools, job alerts, candidate flows, responsive communications, staging/deployment, and cross-market delivery.
Apple: service, training, localization, and the human standards behind a high-trust experience.
EF / Englishtown: hands-on web, mobile, learning-product, CMS, and campaign-system craft at regional scale.
Different companies, different tools. I kept running into the same problem: good people carrying complexity the product or team system should have absorbed.
Operating questions
The projects change. The questions repeat.
What evidence actually changes the decision?
Signals are cheap to collect and expensive to carry. The useful work is turning them into choices a team can compare.
What load should the product absorb?
The best simplification is not just fewer steps. It is moving the right burden away from the person without removing necessary judgment.
How I work
How decisions get made.
Not a new identity. The method behind the work: make complexity clear enough for a team to align and act.
Frame the ambiguous situation
Name the real decision before the artifact: trust, burden, direction, standards, or control.
Virtual Recruiter · HabitatMake signals comparable
Turn feedback, research, metrics, and constraints into options a team can compare.
EMA · Flash FunnelBuild the decision environment
Use maps, roles, rituals, and governance so functions and regions can move together.
Virtual Recruiter · SEEK · Indeed UniversityLeave capability behind
Leave guides, standards, loops, approvals, or recovery paths that others can keep using.
Indeed University · Apple · HabitatSelected work
Three flagship cases carry the main proof.
Each case has a different job: trust and delegation, product synthesis, or decision-load reduction.
What the employer believes
What the system is allowed to do
01 / BurdenRecruiting effort is too highTargeting, qualification, outreach, prioritization.
02 / GuideProduct can narrow choicesMake the next decision easier without hiding why.
03 / DelegateSystem may act only inside clear boundsExplain action, rationale, and recovery path.
04 / ApproveEmployer remains accountableHuman control stays visible at high-trust moments.
Trust contract mapThemed reconstruction
Virtual Recruiter / D.I.F.M.
A high-ambiguity recruiting concept shaped into a clearer product/service direction by making explainable delegation, trust boundaries, explanation moments, and MVP decisions visible.
Employer effort → trust contract → delegated product system
SignalsMessy evidence from many surfaces
- Vision + whiteboard → app-home ambition
- Feedback synthesis → repeated employer friction
- App reviews + churn → trust and usability gaps
- A11y + localization → inclusive rollout constraints
- Interview / agenda → time-sensitive mobile moments
- Career Scout / AI → assistance boundary questions
SynthesisTurn signals into comparable choices
- Cluster recurring themes
- Separate symptom from cause
- Map confidence and evidence gaps
- Expose tradeoffs across teams
- Connect evidence to employer decisions
DecisionsConvert evidence into product calls
- Prioritize reliability and comprehension fixes
- Sequence app-home status and next actions
- Support interview / agenda moments
- Defer unclear AI bets until boundaries exist
- Route dependencies to design-system work
Supported outputsWhat the synthesis helped clarify
- Prioritized: fix confusing mobile moments
- Sequenced: app home → interviews → assistance
- Deferred: broad automation without trust model
- Framed: roadmap discussion around employer decisions
What made it complexDifferent signal typesMobile moment constraintsRegional rollout needsDesign-system dependenciesAI boundary uncertainty
Direction
Mobile employer experience can be discussed as a product-intelligence system: app home explains status, interviews get timely support, inclusive foundations stay explicit, and AI assistance waits for clear employer control.
Signal-to-roadmap synthesis mapThemed reconstruction
Employer Mobile App / EMA
A product intelligence case about synthesizing app feedback, churn, accessibility, rollout, interview workflows, design-system constraints, and AI/Scout opportunities to inform clearer roadmap discussion.
Signals → synthesis → priorities → decision support
Old burden11 pages / 59 clicksReading, typing, context switches, quality worries, and mobile dropoff before enough momentum existed.
Responsibility sortKeep / remove / defer / prefill / explainEach input earns its place by job quality, confidence, timing, or product support.
New modelOne-page direction + checklist supportMake progress feel lighter while keeping the posting complete enough to trust.
Measurement logicDropoffRead burdenType burdenClicksCompletion behavior
Decision-load reduction mapThemed reconstruction
The Flash Funnel
A mobile-first job-posting case about reducing decision load: auditing an 11-page, 59-click legacy funnel, sorting each input by responsibility, and protecting employer momentum without making the posting feel careless.
Decision load → responsibility sorting → mobile momentum → behavior signals
Labs
The same questions now move into AI.
Habitat is my personal operating system and AI lab. It tests a simple rule: agents can help only when memory, tools, approvals, logs, and recovery paths stay visible.
AI operations lab
Habitat / Local-first AI Agent Environment
A local-first AI environment where agents can help, but only inside visible boundaries: memory, tools, approvals, logs, reflections, and recovery paths.
Useful autonomy needs visible interruption, approval, and recovery.
Leadership & capability
A lot of leadership is noticing when the room is stuck.
That has meant training and service localization at Apple, regional design-system governance at SEEK, UX integration at Indeed University, large cross-functional design sprints, mentoring, facilitation, inclusion leadership, and agent systems where human authority stays visible.
Where next
Choose the route by what you want to understand.
WorkCase studies in hiring systems, mobile strategy, automation trust, and decision-load reduction. AboutThe career pattern, operating beliefs, and how I tend to work. LabsHuman-controlled AI systems, product intelligence, and agent governance experiments. ResumeThe conventional version: roles, focus areas, and selected capabilities. ContactFor product systems, AI, design leadership, collaboration, or speaking conversations.