Framing problems, surfacing risks, edge cases, and validating assumptions with LLMs

And what it looked like to work on a design team that intentionally experimented with AI for several months

It’s been a while since I posted here. I was honestly getting a little tired of the “AI is replacing designers” or “the design process is dead” headlines.

When I opened Medium to write this article, I found a draft I’d started back in 2023 about how I was using ChatGPT in my day-to-day as a product designer. I never published it. Looking back, I think I was afraid of how the UX community would react, so I eventually gave up on it.

Finding that draft reminded me how much my workflow has changed since then. So, here’s my second attempt at writing this article. If you’re reading this… I finally hit Publish. 😅

How do I use AI as a product designer?

I've recently started experimenting with AI in my work to speed up the process. I primarily use Claude Cowork and recently started trying out Claude Design. I'd like to share what I liked:

1. Checking the task for completeness. When we're talking about a product task that results in a specific mockup or flow, it's important that the task is described as thoroughly as possible, as the outcome directly depends on the designer's immersion in the context. That's why running the task through AI and challenging the product designer can significantly change the final outcome.

2. Task planning. I often work on large projects where tasks need to be decomposed, otherwise they're quite difficult to approach. Claude is great for breaking these tasks down into stages and subtasks, which become milestones that are understandable for both you and your team.

3. Conducting quick research at the initial stages of the task. Previously, it was necessary to spend a considerable amount of time defining user segments, jobs to be completed, and so on. Now, it's all done in one or two prompts, and you immediately receive a significant amount of research that helps you immerse yourself in the context of the task.

4. Mobbin's MCP integration into Claude. It opens up a wealth of opportunities, for example, a quick and detailed analysis of direct and indirect competitors, finding killer features, analyzing application flow, and suggesting proposals for your own product. This greatly accelerates the initial immersion in tasks because it immediately provides sufficient context for what's happening in the market.

5. Highlighting weaknesses and analyzing what has been accomplished. For example, you're reworking a flow to improve it and asking Claude to analyze it and compare it with competitors' flows on Mobbin. Typically, 30-40% of the insights turn out to be quite useful, while the rest either miss the point of the interface or provide information that's valid outside the context of the product.

6. Finding missed corner cases. When you're designing a completely new section of the interface that hasn't existed before, it really helps you understand whether you've missed anything; the development team will definitely be grateful for it.

I'd say Claude is now a great work assistant and fits perfectly into anyone's workflow. It allows me to significantly speed up previously time-consuming stages, allowing me to focus on what's most important: high-level planning, polishing mockups and UI work, more time on micro-animations, and ensuring the interface is responsive to user interactions.

I shipped 50+ screens as a solo design engineer in 3 months. Here's the exact workflow I use every single day.

Most designers treat AI like a search engine. Ask it a question, get an answer, go back to Figma.

That's not how I use it. AI is my starting point, not my assistant.

Here's my actual daily workflow, step by step:

Step 1: Problem breakdown (Claude Code, 15 min)
Before I open any design tool, I paste the brief into Claude Code and have a conversation. Not "design me a dashboard." Instead:

"Here's a hiring pipeline with 9 stages. A recruiter needs to move 40+ candidates through these stages daily. What are the information hierarchy challenges? What are the edge cases when 3 stages have 20+ candidates each?"

This forces you to think about the real problem before you start pushing pixels. 80% of bad designs happen because the designer skipped this step.

Step 2: Explore in code, not Figma (Cursor, 1-2 hrs)
I scaffold the actual React component in Cursor. Not a mockup. Real code with real data.

Why? Because a Figma mockup with 5 cards looks great. A real component with 47 cards, 3 empty states, and a loading skeleton tells you the truth about your design.

I catch layout problems in 20 minutes that would take 2 days to discover after handoff.

Step 3: Refine the craft (Figma + Code, 1 hr)
This is where I apply the judgment AI can't give you. Does the spacing create the right visual rhythm? Does the hierarchy guide the eye correctly? Does the interaction feel right, or just function correctly?

AI is fast. But it has no taste. This step is where good design becomes great design.

Step 4: Push a PR (30 min)
What I designed IS what ships. Zero handoff gap. No "the developer interpreted it differently." The designer and the engineer are the same person.

Why this matters for you even if you don't code:

You can still use Steps 1 and 3.
Step 1 works for any designer. Paste your brief into Claude or ChatGPT before opening Figma. Ask it to find the edge cases, the conflicts, the scenarios your PM didn't mention. You'll start every project with a better understanding of the problem.

Step 3 is a mindset. AI can generate layouts. It can't tell you why one layout feels trustworthy and another feels cheap. That judgment is your real skill. Protect it.

The designers who will thrive in the next 2 years aren't the ones who avoid AI. They're the ones who use AI for speed and apply human judgment for quality.
Speed is a tool. Taste is a career.

When I use AI in my design work, I’m not outsourcing my thinking. I’m speeding up the parts of the process where speed is actually valuable, generating options, exploring directions, getting a quick read on structure, so I can spend more time on the parts where my judgment matters. That’s a meaningful difference.

I use Claude for research. I’ve trained it over time to understand the tone and voice of the products I work on. When I save a screenshot of something that inspired me, I send it over and it helps me remember why I saved it, what the design is actually doing, and how it connects to whatever I’m building now. That’s not impressive. It’s just useful.

I’ve also given it instructions for UX writing — the tone, the constraints, who the users are. When it gives me copy suggestions, I don’t paste them straight in. I read them against the product’s voice and ask whether they’re actually clear for the user. Sometimes I run tests to find out.

We need to decide:
Is AI here to make our work faster
or to replace the work we do?

I use AI regularly in my workflow —
to explore ideas, speed up tasks, and meet deadlines more efficiently.

And yes — with the right context,
it can even help identify edge cases, bugs, and gaps.

But that’s the key:
AI works best when it’s guided.
It still relies on human understanding of:
• Context
• Dependencies
• Product thinking
• Real-world constraints

Without that, outputs and timelines can look efficient —
but fall short in reality.

AI is a powerful tool to accelerate the way we work.

But building meaningful products still requires
human judgment, collaboration, and experience.

Humans shouldn’t compete with AI —
we should learn how to direct it effectively.

💬 How are you using AI in your workflow today — as support or substitution?

#UXDesign #ProductDesign #UserExperience #ArtificialIntelligence #AIinDesign #SoftwareDevelopment #FutureOfWork #DesignThinking #TechTrends

AI tools won't replace UX Designers.

But designers who ignore AI will be replaced by those who don't.

Here's how I'm using AI in my UX process right now (without losing the human touch):

🔍 Research synthesis Dump interview notes into AI → get themes in minutes, not hours. I still validate every insight but I save hours of grunt work.

✍️ First-draft microcopy AI gives me 10 variations of a button label or error message. I pick, refine, and make it actually sound human.

🗺️ User journey mapping I use AI to spot gaps in a journey I've been staring at too long. Fresh (artificial) eyes catch what tired human ones miss.

🧪 Usability test prep AI helps me draft screener questions and discussion guides 3x faster. More time for actual conversations with users.

The goal isn't to hand over your design thinking. It's to free up more time for your design thinking.

What's one AI tool that's genuinely changed how you work as a designer? 👇

#uxdesign #aiindesign #ai #userexperience #productdesign #designtools

The design process hasn’t changed. How much energy you have left for it has.

Let me say something that might surprise you.

After years of designing at scale — AI-powered tools for 1.6M Walmart associates, global experiences at Adobe, multi-product workflows at Intuit — I don’t think AI has fundamentally changed what good design requires.

You still need to talk to real users. You still need to understand the business. You still need to make judgment calls that no algorithm can make for you.

What AI has changed is everything around the design. The retrieval. The synthesis. The prep work. The 11pm Slack thread archaeology.

And that changes everything about how much of yourself you have left when you sit down to actually create.

...

AI didn’t change what I design. It changed how I design.

Over the past several months, my design process hasn’t just evolved — it’s been fully restructured. My AI tool stack isn’t something I use occasionally; it is an operational layer embedded into my day-to-day and across discovery, ideation, prototyping, validation, and delivery.

This article outlines how I structure projects using AI models — primarily Claude and ChatGPT — alongside specialized tools such as Figma Make and Amplitude.