How to build an AI-augmented design practice by leaning on 2 mindset shifts
Design process isn’t dead — it’s conditional
The claim “the design process is dead” has become trendy in design leadership. But I’d argue it’s a misreading of what’s actually needed: not the death of process, but the ability to shift mindsets based on the problem type. In my experience, that’s where most design teams actually fail.
Here’s what I’ve learned working on pioneering AI features at Google Search, and now building an AI-augmented design practice at Skroutz Marketplace:
Designers aren’t failing because process doesn’t work. They’re failing because they don’t know when to shift their mindset.
You need different thinking for different problems. Most teams treat all problems the same way.
The two mindset shifts
These two mindsets aren’t new. They’ve always existed in design. The question isn’t whether they exist. It’s whether you’re conscious about which one you’re in, and whether you can shift deliberately.
Mindset Shift 1: The Defined Problem
Others have solved these problems before in one way or another, always with space to innovate based on unique product needs. Think traditional e-commerce flows, checkout optimization, accessibility compliance. Here, the Double Diamond design process works beautifully. You discover, converge, develop, deliver. The signal is strong. You can research, iterate, measure.
Mindset Shift 2: The Unprecedented Problem
The technology doesn’t exist yet, or the problem hasn’t been defined yet, or the industry is brand new with novel interaction paradigms. Early-stage startups exploring a new space live here. Teams building entirely new technologies live here. Here, the traditional design process doesn’t just slow you down. It gives you false confidence in the wrong direction. You can’t research something that doesn’t exist. You can’t validate a user need for an experience they can’t imagine yet.
This is where “learning by making” is effective. And it’s also where you hit the boundary of what AI can actually help with. AI excels at surfacing explicit knowledge (patterns, documented rules, existing solutions etc). But in Shift 2, you’re building something that doesn’t have explicit precedent yet. That friction, making mistakes and learning from them, can’t be automated or prompted into existence. It requires craft, taste, and the willingness to be wrong fast.
The problem isn’t that these two mindsets exist. The problem is that most designers don’t shift between them consciously. They try to Shift 1 a Shift 2 problem (over-process it, waste time) or Shift 2 a Shift 1 problem (ship garbage, iterate recklessly). That’s where things break…
The cautionary tale: Google Circle to Search
In 2023, as part of the Google Search team, we explored a new interaction paradigm: what if users could search directly on their screen instead of typing? The technology of multimodal AI, capable of understanding gestures, was emerging. But the user experience didn’t exist anywhere. No competitors had done it. No research could tell us whether users would adopt it.
We had to shift mindsets. This was Shift 2 — unprecedented territory.
The traditional Double Diamond would be worthless. No competitor to analyze, no user need to validate. We had a technology and a hypothesis: help users search from anywhere. So we made fast prototypes, craft decisions driven by intuition and taste, tested with small groups, failed, and learned.
Months of iteration, many hours with executives, many prototypes, moving from Figma to eng prototypes that felt like the real thing, and many conversations about what “feels right”, led to something we believed in. When it launched, it was a success, and now it’s part of Google’s Gemini experience.
We never would have gotten there by running design sprints. We got there by learning by making.
The temptation: the second diamond
In summer 2025, Skroutz doubled down on AI, both externally in our consumer-facing product (new AI-assisted features) and internally across every function. For us designers, the industry narrative was intoxicating. Everyone was talking about agentic design systems where AI could feed into a mature design system and generate UI components automatically.
It sounds efficient. But for our Product Design team at Skroutz, we wanted to take a step back and consider where we could put our most effort.
The first diamond of the Double Diamond is “discover and define”, build the right thing. The second diamond is “develop and deliver”, build the thing right.
For us, building an agentic design system is entirely second-diamond work. It optimizes for speed and consistency in execution, assuming you’ve already nailed the first diamond, that you know what to build and why.
Building a true agentic DS would have required:
- A design system in perfect parity between Figma and code; not just tokens, but every component rule documented and consistent
- Significant back-and-forth alignment work; auditing and refining every component to ensure the AI could reliably replicate your system
- Tackling the hidden complexity; what looks like a working demo on LinkedIn is actually months of hard alignment work no one talks about. Like an iceberg problem, what’s visible is 10%, the real work is below the surface
- A philosophical shift; “systematize first, think later”, deferring creative judgment to pre-written rules
We had invested in a thoughtfully-maintained design system built hand in hand with our engineering team in summer 2024. It was mature because it reflected hard-won decisions about consistency, accessibility, and brand. But maturity isn’t the same as abstraction. You can’t hand off brand judgment to an AI without first understanding why those decisions were made.
So we faced a choice: Should we optimize the second diamond (build the thing right faster) or double down on the first diamond (build the right thing more intentionally)?
We chose the first diamond. We needed to shift our mindset. Let me explain why.
The pivot: from “agentic” to “thought partner”
Instead of following the LinkedIn frenzy and automating component generation, our design and research team at Skroutz asked a simple question: “How can AI augment our skills in our day-to-day?”
In December 2025, we ran a team workshop to answer that exact question. Among many themes — including the agentic DS — one prevailed as the clear winner: our superpowers aren’t in faster execution; they are in better thinking. After that workshop, we formed a small task force of three designers and one researcher to explore how AI could augment our daily workflows.
So we built an AI-enabled design workflow. A thought partner that supports designers from insight gathering through solution feedback. We’ve tested it in different situations, product areas, and project stages. It’s rolling out now to the team, and the impact is already visible: juniors access the same craft lens as seniors, and seniors focus on strategy instead of QA.
This meant staying in the first diamond (build the right thing), focusing on the work that matters: pulling insights from multiple sources, synthesizing them into hypotheses, framing problems rigorously, and getting structured feedback on solutions. Each requires judgment, not automation.
The workflow works like this:
- Insight Gathering: Our AI tool pulls product data from our data warehouse, research libraries, market intelligence, and support feedback, without bias or fatigue. The designer decides what matters.
- Problem Framing: Before sketching, designers articulate goals, scope, key decisions, and success metrics. This creates shared understanding and organizational memory.
- Design Feedback: Structured AI critique checks accessibility, usability heuristics, cognitive load, and anti-patterns. A junior designer now has access to the same lens as a senior one.
- Decision Logging: Every design decision (goal, hypothesis, constraints, outcome) is captured. When designers leave, their context doesn’t.
Breaking the design process into atomic skills
We didn’t throw away the design process. We broke it into atomic skills — discrete, composable capabilities designers can apply exactly when they need them. Instead of monolithic gates like “design review,” we have: copy & tone, accessibility audit, heuristics review, data synthesis, hypothesis framing.
Designers don’t run every skill on every project. They pull in the ones that matter. For a checkout flow (Shift 1), you might run all of them. For an experimental interaction (Shift 2), you might skip some where craft intuition matters more than checklist compliance.
Going one step further, we’re trialing these atomic skills into ready-made workflows, like “Redesign Checkout,” or “Optimize Performance,” or “Explore New Feature”, so designers don’t need to understand every skill. They just run the workflow that matches their challenge. This keeps the skill library modular and manageable without overwhelming the team.
This is what “atomic” means: not just smaller pieces, but pieces you compose deliberately.
Staying in the first diamond: why thought partnership works across mindset shifts
Here’s what surprised us: the thought partner model amplifies judgment, regardless of which mindset shift you’re in.
Whether you’re optimizing a checkout flow (Shift 1: defined problem) or designing an unprecedented interaction (Shift 2: undefined space), the first diamond (building the right thing) is where judgment happens.
In a Shift 1 problem (checkout optimization), the thought partner helps you:
- Avoid confirmation bias in research
- Surface unexpected data from multiple sources (completely irrelevant at first glance). A meeting transcription, a book quote from your knowledge library, a support ticket, triggering unexpected connections.
- Automate the low-value parts of synthesis and QA
- Spend your human judgment on validation: “Is this actually better?”
In a Shift 2 problem (novel AI interaction), the thought partner helps you:
- Structure your thinking despite ambiguity
- Pressure-test assumptions before prototyping
- Iterate faster by getting feedback on multiple directions
- Recognize patterns from analogous domains (even if not direct competitors)
We designed this thought partner to surface the framework, the accessibility rules, heuristic principles, cognitive load patterns. Juniors still do the hard work of developing intuition through repetition and failure. We removed the busywork, the manual synthesis, the checklist tedium, not the friction of learning. That friction is essential. It’s where taste gets built
Because in neither case is the AI replacing your judgment. It’s extending your judgment. And in both cases, you’re staying in the first diamond longer, making sure you’re building the right thing before you optimize the execution.
That’s the real shift. Not “use AI for everything faster.” But “use AI to think more clearly about what’s worth building in the first place.”
The vision: beyond design
Here’s where I think this thinking could go.
If every function — BI, Design, Research, Support, Engineering — built thought partners tailored to their work, what would happen? Each bringing their own data and expertise into a shared layer. A company-wide intelligence system where any team could tap into organizational thinking, regardless of which mindset shift they’re in.
This is territory Ramp has explored with Glass AI — a unified co-pilot layer across design, product, and engineering. It’s an inspiring proof point that the idea has legs.
We’re not there yet at Skroutz. Right now, we’ve proven the concept works but the architecture could scale.
And here is my take: the future isn’t “AI design tools.” It’s organizational intelligence systems that help every team think better, together.
Things We’re Careful About
The apprenticeship problem is real. If companies optimize purely for velocity, automating away the “mediocre early work” that juniors do, they’re not just winning this quarter. They’re outsourcing the training pipeline for tomorrow’s experienced designers.
The friction of foundational work, failing at copy iterations, missing edge cases, getting feedback wrong, is where designers build judgment and taste. You can’t prompt your way into having good instincts. You build them through repetition and failure.
The real competitive moat isn’t shipping faster this quarter. It’s having designers with developed taste and judgment five years from now. That requires protecting the apprenticeship model while using AI to eliminate the busywork, not the learning.
What this means for your team
If you’re a design leader, here’s what to do:
- Diagnose the shift: Ask yourself, which mindset are we in? For Shift 1 challenges (optimization, iteration, consistency), automate the low-value work and free designers to focus on judgment. For Shift 2 challenges (new categories, novel interactions), build rapid prototyping loops, fail fast, test with real users early
- Break your process into atomic skills: Stop thinking about “design review” as a monolithic gate. Compose discrete skills (QA, synthesis, framing, feedback) that designers pull in where they matter.
- Build a thought partner, not a generator: AI’s real value isn’t faster shipping — it’s clearer thinking. Use it to surface data, pressure-test assumptions, democratize craft judgment, and capture why decisions were made. And protect the apprenticeship: the temptation to automate junior work is real, but talent pipelines collapse when you do. Let designers learn through the friction that builds taste.
- The design process exists and it’s conditional on mindset shifts. Match your rigor to your problem. High clarity + low risk = validate fast. Low clarity + high risk = learn by making. Different shifts, same compass.
The age of AI doesn’t mean the end of intentionality in design. It means intentionality is now non-negotiable.
Ioannis Nousis is Head of Product Design at Skroutz, a Greek e-commerce platform. He previously led UX design for Google Search and Google Maps, and earlier built his craft at MOO Print and Transifex. He speaks regularly on design, AI, and building teams — find him at design conferences and workshops across Europe.
Two gears, one compass: designing at velocity while sustaining quality was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.
