- What the research says
- My experience as a designer
- This makes sense, actually
- So what does this mean?
- A closing thought
There's a pattern showing up across industries that nobody talks about enough: the people who gain the most from AI are often the people who needed it most to begin with.
Not the experts. Not the veterans. The beginners.
And if you're already good at your job, you might be surprised to find that AI doesn't do much for you at all.
What the research says
In Co-Intelligence, Wharton professor Ethan Mollick writes bluntly: "In study after study, the people who get the biggest boost from AI are those with the lowest initial ability—it turns poor performers into good performers."
This is backed by one of the more rigorous workplace AI studies we have. In 2023, researchers from Harvard, MIT, and Wharton ran a field experiment with 758 consultants at BCG. AI users completed 12% more tasks, finished 25% faster, and produced work rated over 40% higher in quality. But the gains were not evenly distributed. Lower-performing consultants saw dramatic improvements. Higher performers? Noticeably less so.
AI is a leveller. It narrows the gap between the average and the exceptional. My own experience as a designer backs this up and it's more concrete than it might sound.
My experience as a designer
I've been doing UI/UX design for a decade. I'm fluent in design thinking, comfortable with interaction patterns, and I have a strong sense of how to organise information spatially. When I sit down to design something, the process of making the design is the process of thinking through the design. The two things happen at the same time.
I've tried to change that by bringing AI into my Figma workflow, specifically Figma Make, and more recently Claude Code with the Figma MCP. Here's what actually happens:
Prompting takes time. There's this assumption that prompting is fast and free. But translating a visual idea into words, with enough precision for the AI to do something useful, is genuinely hard. For me, expressing an idea in Figma directly is faster than describing it to an AI.
Design thinking happens while designing. When I work manually, I'm simultaneously making decisions: is the information really necessary for the user, should it be organised horizontally or vertically, should it be disclosed progressively, should this user flow split here or later. Those decisions emerge naturally through the act of designing. AI interrupts that loop. I still have to make all the same decisions, just after the fact.
The output needs revision regardless. AI-generated designs aren't wrong exactly — they're generic. They don't carry my design system, my users' mental models, or the specific constraints of the product. Every screen is a starting point I have to tear apart. Which raises an honest question: is that actually faster than starting from scratch? For me, usually not.
The output needs revision regardless. AI-generated designs are rarely production-ready not just because of misalignments in style and layout, but more importantly, because they are generic. They don't carry my users' mental models, the specific constraints of the product, or my design philosophy. Every screen is a starting point I have to tear apart. Which raises an honest question: is that actually faster than starting from scratch? For me, usually not.
Multi-screen generation doesn't solve this. Tools like Google Stitch and Uizard can generate a full deck of screens at once, which sounds closer to how designers actually work. But the fundamental problem remains: they give you what without telling you why. Even when you feed them context through a prompt—a product description, a reference image, a rough brief—what's missing isn't information, it's intent. The decisions behind the designs aren't transparent, which means there's no reasoning to evaluate, only pixels to accept or redo.
This makes sense, actually
AI at the task level provides a scaffold, helping people do things they couldn't quite do on their own. If you're a junior designer unsure about layout, AI gives you a starting point. If you're a consultant who struggles to structure an analysis, AI gives you a framework.
But if you already have fluency, intuition, and speed, you don't need the scaffold. You're already building on solid ground. The aggregate productivity numbers are impressive. But the distribution of gains is uneven for experienced workers.
So what does this mean?
The next wave of AI integration, if it's going to deliver real value for experienced professionals, can't just be about task automation at the surface level.
What would actually be useful is something deeper. Not AI that generates screens from a prompt, but AI that works within a real design pipeline, holds user research and business constraints in mind simultaneously, applies UX principles consistently across a full product, and documents its reasoning so the designer can evaluate, challenge, and override it with confidence.
That's a genuine system-level collaborator. And it earns its place not by being faster than an expert at isolated outputs, but by operating at a level of complexity no single person can hold in their head at once.
We're not quite there yet. But the direction is clear and it connects to a broader shift I've been thinking about: moving from AI as a task executor to AI as a thinking partner embedded in the design process. I wrote about a related idea in the context of product development here—the argument that treating our best ideas as assumptions rather than requirements is what makes rapid, AI-assisted experimentation actually work.
A closing thought
AI as a great equaliser is a genuinely good thing: fewer skill gaps and more people with a foothold. But it also means that adopting AI tools isn't automatically a competitive advantage for experienced workers. In some cases, AI-mediated workflows can even be inefficient.
The more honest question isn't "are you using AI?" It's: where in your workflow does AI add something you couldn't do better yourself? That's where the real value is.
I'm planning to dig deeper into what that next wave of AI integration could actually look like in practice—what it means to move beyond task automation toward something more systemic. If that sounds interesting, follow along.