What It Means to Be an eBay AI Ambassador
eBay launched a company-wide AI Ambassador program a small group of individual contributors selected across all functions to move AI from experimentation to practical, enterprise-scale impact. I was chosen because I had already done it.
By the time the program launched, I'd already spent 12 months in the work building three AI agents for eBay's People Team that handled benefits questions, financial guidance, and HR case routing for 10,000+ employees. The program gave it a name. The results were already there.
What I actually do as AI Ambassador
- ▸ Biweekly People Team enablement sessions hands-on AI sessions with HR staff covering prompting, agent workflows, and practical tools they can use the next day
- ▸ Monthly People Org AI best practices shares broader sessions for the wider People organization covering what's working, what's not, and what's new in enterprise AI
- ▸ AI Transformation Workshops outside the People Team ran a full session for eBay's Global Markets Success team introducing agentic workflows, GenAI tools, and take-home playbooks
- ▸ Identify and prioritize high-impact AI workflow opportunities across eBay functions
- ▸ Pilot enterprise AI tools and build lightweight frameworks other teams can pick up without needing engineering help
- ▸ Turn what works in one team into patterns other teams can reuse so we're not solving the same problem six times
- ▸ Represent the people side of AI in steering discussions what employees actually need versus what sounds impressive in a demo
Five Principles I Won't Compromise On
After over 12 years in product design and a year leading AI integration, these are the non-negotiables I bring to every AI conversation.
I've seen teams hand research synthesis or copy generation to AI and walk away. The output looks complete but the insight is shallow - because AI doesn't know what question actually matters yet. My rule: AI gets involved after the designer has formed an initial hypothesis. It sharpens thinking, not substitutes it.
When I worked on the Magical Listings price suggestion, a single number recommendation felt arbitrary. Adding a "Why this price?" explanation panel - with confidence range and comparable sales anchors - changed adoption by 34% in A/B testing. Users don't just need the right answer. They need to understand it before they'll act on it.
AI tools will always surface speed as the headline benefit - and it's real. But I consistently push teams to also measure: did the quality go up? Did the team learn something new? Did users trust the output? A 33% reduction in research time means nothing if the insights are worse. The scorecard has to be broader.
This is the principle I push back on most in product reviews. Engineers often frame AI decisions as too complex to surface. My counter: if we can't design a clear explanation for the user, we haven't understood the AI's reasoning well enough ourselves. Explainability starts in the design brief, not the legal disclaimer.
I treat the prompts I design with the same rigor as the interfaces I design. Length constraints, tone parameters, specificity injections, guardrails - these are not engineering concerns. They are UX concerns, because they directly determine what the user sees and how they respond to it. I co-own prompts with engineering and I iterate on them through usability testing.
Helping People Actually Use AI
Building agents is one part of the job. The other part is making sure people know how to use them, trust them, and keep using them after the launch excitement wears off. That takes ongoing work, not a one-time rollout.
Every two weeks I run a hands-on session with the People Team. We work through real prompts, test new tools against actual HR workflows, and talk through what's working and what isn't. No slides-only presentations. The goal is always something people can use the next morning.
Topics covered: prompt writing for HR use cases, how to evaluate AI outputs critically, when to trust the agent vs. escalate to a human, new tools as they come out.
Once a month I run a broader session for the full People organization. This one's less hands-on and more about what's happening across the field: what AI tools are actually delivering results, what teams are struggling with, and where the next opportunities are. It keeps everyone connected to the same picture even if they're not in the biweekly sessions.
Format: short demos, real examples from the team, open Q&A. Designed to fit in 45 minutes and leave people with at least one thing to try.
Part of the AI Ambassador role is taking what works inside the People Team and making it available to other parts of eBay. I ran a full AI Transformation Workshop for the Global Markets Success team, covering agentic workflows, hands-on GenAI tool use, and a take-home playbook they could actually run with. The goal with external workshops is always the same: leave the team better equipped than when you walked in, not just inspired.
Approach: workflow-first (real problems, not toy examples), hands-on practice during the session, playbook with tested prompts and tool shortcuts to take away.
AI Agents for eBay's People Team
The most significant AI transformation work I've led wasn't on the product side it was internal. Embedded in eBay's People Team (HR), I served as both Product Manager and UX Lead for a full AI workflow redesign, bridging the People Team and Engineering (IT Services & Solutions) to deploy three agentic AI systems that changed how 10,000+ eBay employees interact with HR.
Three Agentic Workflows, Three Workflow Problems Solved
AI Across Every Phase of the Design Process
Systematically embedded AI across every phase of eBay's design process research, ideation, prototyping, testing. Not just the shiny visible phases. Each step measured.
Each phase had a different integration challenge. In research, the question was: how do we use AI to surface patterns without losing the texture of individual user stories? In design, it was: how do we use AI generation without letting it homogenize our output? In testing, it was: how do we use automated analysis without outsourcing our judgment about what actually matters?
The framework I introduced: Human-Centered Opportunity Sizing
Before any AI tool gets introduced to a workflow, I run it through three questions: What human need does it serve? What human judgment does it preserve? What happens when it gets it wrong? Any tool that can't answer all three doesn't get recommended. This framework is now used by the broader design team at eBay when evaluating AI tools.
The AI Toolkit I Built and Teach
12 months of real project pilots not vendor demos. Every tool was tested on live work before being recommended to the team.
Each tool was piloted on a real project before being recommended to the wider team. I documented what worked, what didn't, and under what conditions each tool adds genuine value vs. creates noise. That documentation is now the team's shared AI toolkit.
What Changed, and How We Know
Every number below was measured before and after AI integration, against a control or baseline. I was part of designing the measurement approach - because if you don't define what success looks like before you ship, you'll rationalize whatever you get after.
11 AI & UX Certifications
I pursued these certifications deliberately - not to collect badges, but to build a rigorous foundation for the recommendations I make to the team. Each one informed a specific decision or workflow change at eBay.
What Leading AI Design Has Taught Me
Being the AI Ambassador means being the person who has to make the case - to skeptical stakeholders, to excited engineers, and to designers who are afraid their work is being devalued. I've had all three conversations. Here's what I've learned.
The best AI tools make designers more curious, not less
When AI handles the volume work - clustering, synthesis, layout generation - it frees the designer to focus on the questions that actually matter. That's the right direction. Tools that make you stop asking questions are a warning sign.
Responsible AI advocacy requires saying no
I've killed features that looked great in demos because they failed the "what happens when it's wrong" test. That credibility - of being the person who holds the quality bar - is what makes the yes decisions land with weight.
The real work is cultural, not technical
Getting a team to use Figma AI is easy. Getting a team to trust AI outputs, question them intelligently, and iterate on prompts the same way they iterate on designs - that takes months of modeling the behavior yourself.
AI doesn't change what good design is - it changes how fast you get there
Good design is still grounded in human understanding, clear thinking, and honest tradeoffs. AI accelerates the path to those outcomes. It doesn't change the destination.
To discuss AI design leadership, the frameworks I've built, or how I approach responsible AI integration at scale, connect with me on LinkedIn / Nancy-UX.