Sienna Chen · Design Approach
Three lenses I use across projects in the age of AI.
Jenny Wen, former Director of Design at Figma, left her role to return to IC work at Anthropic. Andrej Karpathy, OpenAI co-founder and the person who coined "agentic engineering," named what's actually happening to our field.
Two things matter more than ever as a designer: engineering craft, and the judgment to direct it.
One of the biggest mindset shifts I got from Parsons was learning when to step back from the problem entirely.
Most people enter a business challenge at the subject level — optimizing a product, fixing a process, refining a feature. However, the first question I ask is whether we are looking at a symptom or at the system producing it. This is where strategic design separates from UX design; it is a judgment call driven by systemic thinking.
Much of this thinking was shaped by my professor Raz Godelnik, whose work on strategic design continues to influence how I approach complex problems.
Two frameworks I use to read the strategic environment before any design work begins.
Before I define a design direction, I need to know which strategic environment the business actually operates in — predictable, adaptive, harsh, or visionary — because each demands fundamentally different design logic.
BerylElitesX
The Palette revealed a mismatch: Adaptive environment, Classical design logic. That reframe reshaped the strategic brief.
My practice is to establish H1 and H3 before engaging H2. Without those anchors, the middle horizon drifts into optimism or paralysis. Roger Martin's question sharpens the work further: "What would have to be true for this transition to be possible?"
Google Maps Sustainability
Work sits in H2+. H1 = existing eco-certification infrastructure. H3 = a world where sustainable choices are effortless and invisible. The design question only becomes precise once both ends are established.
Referenced · Roger Martin's Strategic Choice Structuring Process
Where strategic thinking connects directly to what gets built, how it's positioned, and how it enters the market.
An impact vs. effort matrix turns team debate into a decision by making judgment criteria explicit — what to build now, what to watch, and what to drop.
Bayer
After framing four opportunity directions from research, a core team session aligned on prioritization and stakeholder buy-in in one room.
I combine Porter's Five Forces with the BCG Strategy Palette to analyze industry power dynamics and strategic environment simultaneously.
Parkonomy
The analysis revealed the competitive moat wasn't in the parking transaction — it was in the data layer. That reframing changed the entire product direction.
GTM is a product decision. Entry point determines which features matter, which partnerships are essential, and which users get prioritized first.
Parkonomy
B2B entry targeting property managers over individual drivers — validated with a letter of intent from one of China's largest real estate firms.
Most designers today use AI in some form. That part is no longer interesting.
The harder question is how to meaningfully integrate AI into a design process without losing the judgment that makes research worth doing. The whole industry is still figuring that out in this extremely fast-changing environment.
However, I am still able to contribute some ideas based on my experience and continuous thinking.
In the podcast, Jenny mentioned that the old linear design gospel — research, diverge, converge, prototype, repeat — can no longer hold its shape when engineering moves this fast. Something has to give.
But what should give, and by how much. My answer is that it depends on how much ambiguity the project carries before the first prototype exists. I assess this across three dimensions. User diversity, Feature complexity, Stakes.
The Bayer project sat high on all three. That's why we spent four weeks on research and opportunity framing before anything got built.
For high-nuance projects, I follow the original design thinking structure. Primary and secondary research run before problem definition is locked. Prototyping only begins once the problem space is understood, but at that stage AI accelerates everything.
For low-nuance projects, the process compresses. Primary research establishes directional confidence, then the work moves straight into an AI-assisted design innovation sprint. Rapid problem definition, then immediately into AI prototyping.
The signal to stop is the same in both cases. When a working prototype will tell you more than another round of research, ship the prototype.
This is where most research stops too early. Dan Hill's concept of dark matter describes the invisible forces surrounding any design intervention — organisational culture, power structures, policy constraints, institutional inertia. The product is visible matter. Everything determining whether it can survive and scale is dark matter.
User research tells you if people want the thing. Dark matter research tells you if the system will allow it. A solution that users love but contradicts an organisation's incentive structure will not land. Identifying that early is not pessimism. It is the difference between research that informs a decision and research that only describes a problem.
The methods toolkit below remains the same. What changes is why and when each method gets used.
I work across qualitative and quantitative methods. Selection from the methods toolkit depends on the project.
What do we actually need to learn? What decisions will this research inform?
Method selection follows the question.
Interviews, observation, surveys — capturing raw data without interpretation.
Affinity mapping, thematic coding, insight laddering. The goal: compress raw observations into patterns, and patterns into insights.
Research ends when a decision gets made. Make sure those insights have a receiver.
Figma to code. Code to Figma. Code to code. The entry point shifts with every project. I move between tools fluidly.
The Tool Constellation
Any node can be the entry point. Arrows can run in either direction.
© Sienna Chen. Original diagram — do not reproduce without permission.
After research, before code. Manual or AI-assisted depends on project scope. Here's how you decide →
Design literacy = precise prompts. Can't describe the effect? Prototype it first, then show Claude what to refine.
Pull real CSS from DevTools or Stitch, pass to Claude. Grounded output — not generic AI defaults.
The AI-First Tool Stack
4.1 · What Kind of Designer I Am
In the podcast, Jenny Wen described three kinds of designers the industry tends to value in the age of AI — Strong Generalist, Deep Specialist, and Crack New Grad. I'm a hybrid of the first and the third. I call myself a full-stack designer.
That's me
Operates across the full problem stack — from strategic framing to shipped product. I'm also very resilient and adapt to project phases.
At Beryl Consulting and Parkonomy, I operated in this full-stack capacity from day one. A personal project, a sustainability service redesign for Google Maps, is currently in progress. At Bayer, I have deepened my user research and product strategy skills through close team collaboration.
4.2 · My Commitment