AI changed who can create software. The interesting challenge is deciding what's worth building.

Everyone's building now. The walls between product design, product management, and engineering are coming down, and that's genuinely exciting. More people building means more ideas getting tested, faster.

Need an authentication flow? A well-written description from a PM or designer should be enough to get it done. Nobody should be mocking up a login screen for the hundredth time. That era is over, and good riddance.

Where it gets interesting is at the edges. AI handles well-defined, straightforward tasks across disciplines remarkably well. The assumption that follows — that this scales to every kind of problem — is where things start to break down.

Fast Doesn't Mean Good

Can a PM prototype, code, and ship a feature using AI? Yes. Can a developer plan a product and generate a polished interface? Technically, yes. Can a designer build and deploy without engineering support? Absolutely.

But being able to do something and doing it well are very different things. Each of these roles carries years of accumulated knowledge about best practices, edge cases, and the subtle details that separate functional from good. AI is an incredible accelerator, but it accelerates in the direction you point it. If you don't have deep expertise in the area you're working in, you won't know when the output is wrong — or worse, when it's almost right.

AI can handle a clean, well-scoped task with clear inputs and expected outputs. But the moment complexity enters — and it always does — you need people who understand their domain deeply enough to know what to tweak, what to push back on, and what to throw out entirely. AI makes each person on a team faster. It doesn't make them unnecessary.

Start With Something, Then Make It Better

That said, one of the most useful things about AI is that it eliminates the blank canvas problem. I've done a lot of 0-to-1 design over the years, and I can tell you — it's almost always easier to react to an existing thing than to conjure something from nothing.

Generate a rough version. Put it in front of yourself or your team. Start reacting. What works? What's wrong? What's missing? That process of refinement is often faster and more productive than staring at an empty Figma file trying to imagine the perfect solution.

But — and this is the part people skip — that first draft needs someone who actually knows what "better" looks like. Which brings me to the real problem.

What AI Gets Wrong About Interfaces

I ran into this recently. I was working on an AI-coded product that needed API connections, content calendars, master settings, and a number of other features to function. The AI built it all. Every feature was there. And the result was terrible.

The page was long. It had a pile of tabs. Every feature lived in its own section, neatly separated — the way a developer might organize a codebase. Clean architecture, bad experience. If you were a new user trying to set up automation for the first time, you'd land on this page and have no idea where to start. You'd click around, open tabs, lose your place, and eventually either figure it out through trial and error or give up and ask support. The AI was thinking about the product as a collection of features. It wasn't thinking about it as something a person has to sit down and use.

I fixed it by asking a different question: not "how do I organize these features" but "what does someone need to see and do, and in what order?" Four tabs went away. The main page was condensed to fit mostly above the fold. The click-to-completion path for API setup dropped by 34%.

This is the gap. AI is exceptional at addition. You describe a feature, it builds it. You describe another, it adds it. But it has no instinct for subtraction — for knowing when the interface has enough, when another tab or section or option is making the product worse, not better. It doesn't optimize for simplicity because simplicity isn't about what you build. It's about what you choose not to.

Design and Product Are Converging. That's Not New.

Product design and product management have been moving toward each other for a while — AI is just accelerating it. To be effective as a product designer today, you need to understand business context, strategic priorities, and technical constraints alongside the traditional UX skill set. You can't design in a vacuum and expect the output to survive contact with reality.

But the roles aren't collapsing into one. The product manager owns the strategic and business layer — what to build, why it matters, how it fits the bigger picture. The product designer owns the customer experience — how it works, how it feels, whether it actually serves the person using it. Those are different jobs that require different instincts. When I'm deciding whether a feature should exist at all, that's product thinking. When I'm deciding how that feature should behave so a user never has to read a tooltip, that's design thinking. Both happened in the same meeting, sometimes in the same sentence — but they're pulling from different skill sets, and both get better when someone with real depth is doing them.

The Designer's Job Is Shifting From Production to Judgment

What's becoming more important — not less — is the need for someone who can hold the entire product in their head alongside a deep understanding of the customer, and make decisions from that position.

Someone who looks at a fully functional AI-generated interface and knows it needs to lose half its features. Someone who can feel when an interaction is three clicks too many, even if every individual screen looks fine. Someone who asks "does this need to exist?" before asking "how should this look?"

That's the job now. Not pushing pixels — deciding what the pixels should be doing in the first place. And that kind of judgment doesn't come from prompting AI. It comes from years of watching real people use real products and understanding why they struggle.

Design Systems Are the Multiplier

One more thing I've learned from working extensively with AI-generated products: a strong design system changes everything.

Without one, AI outputs are inconsistent — different spacing, different patterns, different interaction models on every page. You spend more time cleaning up the output than you saved generating it. With a well-built design system, AI has constraints to work within. Established patterns to follow, component libraries to pull from, rules that keep the output coherent. The design system becomes the guardrails that AI needs but can't create for itself.

This is another reason the designer's role gets more important, not less. Someone has to build and maintain those systems. Someone has to define the patterns that keep a product consistent as AI generates more and more of the individual pieces. The person setting the constraints is more valuable than the person working within them.

As AI makes building easier, the question shifts. It's no longer "can we build this?" — it's "should we?"

That's a design problem. And it's one that gets harder, not easier, the more powerful the tools become.