From sceptic to daily user

When GPT-3 landed I was curious but unconvinced. I'd played with creative AI tools before — they felt like toys, impressive demos that didn't survive contact with real constraints. Real product design involves systems thinking, edge cases, brand guidelines, accessibility requirements, technical feasibility. A tool that generates pretty pictures didn't solve my problems.

Then two things happened. First, the models got dramatically better at reasoning — not just generating, but understanding context and constraints. Second, I started treating AI not as a replacement for design skills, but as an amplifier for the thinking I was already doing. That shift changed everything.

Today, AI is embedded in nearly every part of my design process. Not as autopilot — as a copilot. The distinction matters.

AI doesn't replace design thinking. It compresses the distance between having an idea and testing whether it works.

Where AI fits in my process

I don't use AI for everything. I use it where it creates genuine leverage — where it takes something that used to cost hours and makes it cost minutes, without sacrificing quality. Here's where it's made the biggest difference:

Research synthesis

Turning hours of user interview notes, support tickets, and analytics data into structured insights. AI helps me surface patterns across large datasets far faster than manual affinity mapping — though I still validate the patterns myself.

Content and copy exploration

Generating multiple UX copy variants for error states, onboarding flows, empty states. Instead of agonising over one version, I generate twenty, pick the best three, and refine. The iteration cycle collapses from days to minutes.

Rapid prototyping

Using AI coding tools to build functional prototypes directly from design concepts. I can test interactions, validate flows, and demo ideas to stakeholders with working code — not static mockups — in a fraction of the time.

System documentation

Drafting component specs, design tokens documentation, and API contracts. AI handles the tedious structure; I focus on the design decisions. What used to take a full day now takes an hour of editing and refinement.

My AI toolkit

I'm tool-agnostic — I use whatever works best for the task. But here's my current stack and how each piece fits:

01

Claude (Anthropic)

My primary thinking partner. I use it for design critique, exploring architectural options for features, writing and refining documentation, analysing research data, and working through complex UX logic. The extended thinking models are particularly useful for systems-level design problems where you need to hold many constraints simultaneously.

02

Kiro / Cursor

AI-native development environments that let me build functional prototypes and tools directly. I've built internal dashboards, prototype apps, and even this portfolio site using AI-assisted coding. It's collapsed the gap between "design concept" and "working thing people can test."

03

Lovable / Bolt

For rapid full-stack prototyping when I want to go from idea to deployed testable app in hours rather than weeks. Perfect for validating concepts with real users before committing engineering resources to a full build.

04

Paper (Dropbox)

AI-powered design tool that bridges the gap between Figma and code. I use it for exploring visual directions quickly and generating production-ready assets without the overhead of full design tool workflows.

05

Midjourney / Image generation

For mood boards, placeholder imagery during early concepts, and exploring visual directions. Not for final assets, but invaluable for communicating intent during ideation when stock photography falls flat.

How it changed my role as a leader

The impact on my individual work was significant. But the impact on how I lead the team has been even bigger.

Raising the floor

AI tools have made our junior designers more capable, faster. Someone in their first year can now produce research summaries, component documentation, and content variations at a quality level that used to require three years of experience. This doesn't make seniority irrelevant — it frees senior people to focus on the judgment calls, the systems thinking, and the stakeholder alignment that genuinely require experience.

Faster feedback loops

When a designer can prototype something functional in an afternoon instead of a week, we test ideas sooner. We fail faster. We learn quicker. Our average time from concept to first user feedback has dropped significantly since we started incorporating AI tools into our workflow.

Redefining "done"

The bar for what constitutes a design deliverable has risen. When you can produce a working prototype instead of a static mockup for roughly the same effort, stakeholders (rightly) start expecting working prototypes. This is a good thing — it means we're testing with higher fidelity and catching more issues before engineering starts building.

The designers who thrive in an AI-augmented world aren't the ones who resist the tools. They're the ones who understand what problems are worth solving — and use every available tool to solve them faster.

What AI can't do (yet)

For all the acceleration, there are things AI still can't replace. Being clear about these boundaries is important — both for managing expectations and for understanding where human designers remain essential:

  • Taste and judgment. AI can generate options. It can't tell you which one is right for your users, your brand, and your technical constraints simultaneously. That requires accumulated experience and contextual knowledge.
  • Political navigation. Shipping a product in a large organisation is as much about alignment, persuasion, and timing as it is about design quality. AI doesn't sit in the room and read the body language.
  • Genuine empathy. Understanding a user's frustration by watching them struggle with a terminal in a noisy betting shop — that embodied understanding can't be simulated.
  • System-level accountability. When something goes wrong, someone needs to own the decision. AI is a tool; designers are responsible.
  • The "why" behind the "what." AI can produce a solution. It can't interrogate whether you're solving the right problem in the first place.

My mental model for AI in design

I think about AI adoption through a simple lens: amplification, not automation.

Automation means removing the human from the loop. Amplification means making the human more effective while they stay in the loop. For creative and strategic work — which product design fundamentally is — amplification is the right frame.

Here's how that plays out practically:

  • Use AI to generate options, then apply human judgment to select and refine
  • Use AI to handle the tedious parts (documentation, formatting, repetitive patterns) so you can focus on the interesting parts (strategy, edge cases, delight)
  • Use AI to challenge your thinking — ask it to poke holes in your approach, suggest alternatives you haven't considered, or explain tradeoffs you might have missed
  • Never ship AI output without human review. The speed is in the generation; the quality is in the curation.

Where this goes next

I'm genuinely excited about what's coming. A few areas I'm watching closely and experimenting with:

AI-native design tools

Tools where AI isn't bolted on but is the fundamental interaction model. Where you describe intent and the tool proposes design solutions you can refine. We're seeing early versions of this now — in two years it'll be the norm.

Personalised interfaces

Products that adapt their UX per user in real-time, powered by AI understanding of context and behaviour. Designers will shift from designing one interface to designing the system that generates many interfaces.

Agents in the design process

AI agents that can autonomously perform research, generate variations, run accessibility audits, and prepare handoff documentation — all while you focus on creative direction and strategic decisions.

Human–AI collaboration patterns

New interaction patterns for how humans and AI work together — not just in products we design, but in the tools we use. The best AI experiences will feel like a skilled collaborator, not a command-line interface.

Advice for designers navigating this shift

If you're a designer wondering how to approach AI, here's what I'd say based on my own journey:

  1. Start with your actual workflow. Don't learn AI tools in the abstract. Pick the most tedious part of your current process and see if AI can compress it. You'll immediately see where the value is.
  2. Invest in prompt craft. The quality of AI output is directly proportional to the quality of your input. Learning to articulate design problems clearly, with context and constraints, is a skill worth developing.
  3. Stay critical. AI is confident even when it's wrong. Build the habit of questioning output, checking facts, and testing assumptions. Blind trust in AI output will burn you.
  4. Double down on what AI can't do. Strategic thinking, user empathy, stakeholder navigation, systems design — these become more valuable as execution-level tasks get cheaper.
  5. Share what you learn. The field is moving fast. The designers who openly share their experiments, failures, and discoveries help the whole community level up.
The question isn't whether AI will change product design. It already has. The question is whether you'll shape how it changes your practice, or let it happen to you.

I'm still early in this journey — we all are. But I'm convinced that the designers who lean in, stay curious, and maintain their critical eye will do extraordinary work in the years ahead. The tools have never been more powerful. The challenge is the same as it's always been: understand the problem deeply, and create something that makes people's lives better.