The process is the distinction

The asymmetry between idea and execution is where designers once lived. A product or flow could be imagined, but only a subset of the workforce could think through the problems, stress test it, finalize, and ship it. Designers existed in that gap between "I have an idea" and "here's the output". This gap has more or less closed.

For most of product design’s history, anyone could imagine an app or flow, but few people could think through the problems to ship something real. Designers lived in that gap between “I have an idea” and “here’s the output.” This gap has more or less closed.

Nowadays, anyone with a half-baked idea and access to a free tiered LLM can generate an interface. Non-technical founders can describe an application and watch UI take shape on a screen. PMs can be briefed on a project and spin up a working prototype before lunch. Friends in my own circle have had the want to build something and suddenly are given the ability to execute. However, we’re missing an important distinction: a working prototype is not a designed product. Generation is not design.

Generation is not design.

There is a sharper line between something that can be produced and something that is ready to exist in the world as a finished product. Renderable output means the pixels are there, arranged in a way that appears polished. The visual artifact gets paraded through a demo, or posted to a Dribbble feed. Shippable, however, means it can be held up under real pressure. There are empty states and error states, and the copy is written for a specific audience. The flow can handle people across different use cases, and the visual language stays coherent on every screen. The “cool thing” AI made, or a polished demo posted on Twitter, does not survive real users. The startup founder generating their MVP in a weekend doesn’t notice this dissonance. The designer working on the product day in and day out, and the consumer using the interface, notice it instantly.

The tool doesn’t separate the people making shippable work from those generating impressive demos. Solely using Figma in your workflow does not inherently make you a more “real” designer. The difference is in the process, and process compounds into quality. AI does not replace the design process, but instead has exposed which parts people have already been skipping.

AI does not replace the design process, but instead has exposed which parts people have already been skipping.

Leaning back into the fundamentals of design (the parts that felt slow and unsexy) made the output categorically cleaner. The work felt more intentional and grounded in reality. This improvement does not come from AI tools getting better. Rather, it comes from the designer becoming more disciplined about everything that happens before the output. I noticed this most recently when building my own portfolio. I relied heavily on AI to move fast, but early drafts didn’t give me a distinctive voice; they were obviously generated and forgettable. So, I went back to look at designers I admired and the moves they made to distinguish themselves. I noticed how Rita Wang treated her case study covers, and started paying attention to which designers used type as a primary visual element instead of imagery. The work changed because I finally knew what I was looking for.

These design pillars kept proving their worth throughout the build, keeping me aligned before handing anything over to AI:

Inspiration and competitive analysis calibrate taste. Before touching anything generative, observing the landscape is essential. What does the best version of this thing already look like in this category and adjacent ones? Where is the unmet space? AI is incredibly good at giving you an average answer for a class of problems. But if you don’t know the shape of that space, it’s hard to distinguish if the return is exceptional or merely plausible.

Low fidelity work calibrates structure. Wireframes and early mockups force decisions to be made before AI ever generates a screen. The hierarchy is already decided, as are primary actions; wireframes answer what gets surfaced and hidden. If the structure isn’t right at the low fidelity mark, then the high fidelity version will amplify these issues.

Visual direction and design system tokens calibrate coherence. Within the enterprise design system at Excellis Interactive, the projects must live within or extend the established patterns. Generating screens inside that framework keeps them coherent because the design decisions are already made. Type scaling, color semantics, spacing rhythm, and component conventions force AI to fill an already established structure instead of inventing one. Stepping outside this risks immediate inconsistency.

Understanding the end user calibrates language and trust. The tone of error states and the framing of onboarding steps are parts of a product that build trust. Tonally generic copy causes issues for an interface rooted in reality, because real users aren’t generic. Language representing a customer service rep in an enterprise platform is vastly different from what a consumer needs from a retail app. Reps must have short, jargon-tolerating copy they’ll see a thousand times a week. Consumers need warm, plain language because they haven’t memorized your product and need to be reminded of things more than once. AI will often give a vague retelling, collapsing these two voices into a well meaning middle. That voice is wrong for both.

The fair pushback is that AI tools are getting better within these areas. They’ve gotten better at scanning landscapes and writing for a specific voice. So the question becomes, won’t they eventually absorb the process entirely? The truth is they’ll get closer. But the parts of the process that actually matter aren’t renderable artifacts; they’re the decisions underneath the screens. AI can produce a competitive analysis but can it tell you what references actually matter for a product you’re building right now, with this team, with these users, against this constraint? It can produce a wireframe but it can’t decide what should be most prominent on a screen when the user only has 40 seconds and three conflicting priorities. AI gives you the average answer for a class of problems. Designers make judgement calls that close the gap between the average answer and the right answer for this specific moment. That gap doesn’t scale by prompt.

AI gives you the average answer for a class of problems. Designers make judgement calls that close the gap between the average answer and the right answer for this specific moment.

When execution was the bottleneck, sloppy structure and inconsistent visuals were brushed off as unimportant, because generating something was impressive enough on its own. Now that generation has become commonplace, we have to sift through the slop to see if any of the underlying decisions were actually made well. The pillars have only gotten more important, because they’re the only thing standing between a half-formed idea and something that deserves to exist.