What Happens When Designers Stop Designing and Start Prompting?
For as long as most of us have worked in digital product design, the job has looked roughly the same. You research, you sketch, you design screens, you hand them over, and someone builds the thing. The designing part and the building part have always been separate.
Now a new generation of AI tools is collapsing that separation entirely. Instead of drawing what you want, you describe what you want, and the AI builds a working product directly. No screens. No handoff. No gap between the idea and the thing you can actually click through.
Which raises a question worth taking seriously: if designers stop designing in the traditional sense and start prompting instead, what actually changes? Is the work better, faster, and more connected to real outcomes? Or does something important get lost when the drawing board disappears?
This article does not have a definitive answer, because nobody does yet. What it has is an honest look at what is happening, what we know so far, and what I am personally doing to find out.
What Is Vibe Coding? A Plain Language Definition
Vibe coding is a term coined by Andrej Karpathy, co-founder of OpenAI and former Head of AI at Tesla, in February 2025. It describes a new way of building software where instead of writing code yourself, you describe what you want to build in plain language and an AI tool builds it for you.
For example, instead of spending days designing screens in a tool and then passing those screens to a developer to build, you simply say: "Build me a dashboard where users can see their appointment history and book a new appointment." The AI generates a working, interactive product, not a drawing of a product, but the actual thing you can click through and test.
The result is that the prototype and the product become the same thing. There is no separate simulation step in between.
The Traditional Way of Building Products (And Why It Has Always Been Slow)
To understand why this shift matters, it helps to understand what the traditional process looks like.
For most of the last decade, building a digital product meant moving through a long sequence of steps. A team would first research what users need. Then they would sketch ideas. Then a designer would build detailed visual mockups in a tool like Figma, which is essentially a digital drawing board. Then someone would build a clickable prototype to simulate how the product would feel. Then that simulation would be handed to engineers who would interpret it and write actual code. Then a quality team would test it. Then it would launch.
Every step in this sequence is valuable. But every step is also a place where time gets spent, where information gets lost in translation, and where the gap between what was designed and what was eventually built tends to widen.
The new approach attempts to compress or remove several of those steps entirely.
Who Is Talking About This Shift, and Why It Matters
In January 2025, Paul Graham, the co-founder of Y Combinator, the world's most influential startup accelerator, posted a single tweet. He said he had spoken to the CEO of a moderately large technology company who told him their team had replaced Figma with Replit, an AI-powered development platform. Graham admitted the comparison surprised him because, in his words, he did not even think of them as being in the same business. But the CEO explained that Replit had become so effective at generating working applications that their team now went straight to a prototype without the design step in between.
That single observation set off a debate across the design and technology community that is still unresolved today.
Sam Altman, CEO of OpenAI, has publicly predicted that software development will look fundamentally different within years. Karpathy's original vibe coding post received millions of views within days of being published. By the end of 2025, Collins English Dictionary had named vibe coding a candidate for Word of the Year, and it had its own Wikipedia article.
These are not small signals. When a concept moves this fast from a single post to mainstream cultural recognition, something real is happening.
The Tools Driving This Change
Several AI-powered tools are at the center of this shift. Each works slightly differently and suits different use cases.
Replit is a cloud-based platform where you can build, test, and launch a full product entirely through AI prompts. It is particularly strong for internal tools and products that need a built-in database. Replit raised 400 million dollars at a 9 billion dollar valuation in early 2026, tripling its value in a short period.
v0 by Vercel is well suited to building polished user interfaces, especially for teams already using Vercel's infrastructure for web deployment.
Lovable is designed for people who are not developers. You describe what you want, and it produces a finished, deployable product. It crossed 2.3 million users and 100 million dollars in annual revenue by early 2026.
Bolt is considered the most versatile of the group, handling both front-end design and back-end logic with speed.
Figma Make, launched by Figma in May 2025, is the incumbent's response. It sits inside the existing Figma environment, which designers already use daily, and converts designs directly into working code using Claude as its underlying AI.
The fact that Figma itself launched an AI-first product is the clearest signal that even the leading design tools company sees where this is heading.
The Real Tradeoff Nobody Is Talking About Honestly
Here is where most articles on this topic pick a side too quickly. The honest position is more complicated.
The case for the new approach is genuinely compelling. Speed increases dramatically. A product that would have taken weeks to move from idea to testable prototype can now be built in hours. Teams can test more ideas, discard the ones that do not work, and iterate faster. According to research from 2024, employees using AI tools report an average productivity boost of 40 percent, with design and development workflows seeing even larger gains.
But there is a question underneath all of this that deserves more attention.
When you remove the slow, deliberate process of designing something carefully, are you also removing the part where the thinking gets better?
Designing in a traditional tool is not just about drawing screens. It is about arguing with yourself about what a user actually needs. It is about the whiteboard session where three people disagree and eventually land on something smarter than any of them started with. It is about the revision that happens because someone pointed out an edge case you had not considered. That process is slow. But the slowness is doing something useful.
When you go straight to a working product using AI, you may be moving faster but thinking less carefully. Or you may not be. Nobody has a clean answer to this yet. The research on quality outcomes, not just speed outcomes, is still thin.
What We Know for Certain, and What We Do Not
What we know:
AI coding tools have reached a level of capability that makes them genuinely useful for building real products, not just demos. Adoption is accelerating across companies of every size. According to data from 2025, GitHub Copilot, an AI coding assistant, has been adopted by 90 percent of Fortune 100 companies. At one major technology firm, all 40,000 engineers now use AI-assisted coding tools as standard practice. Another company reported 55 percent faster task completion after rolling out AI tools to 50,000 developers.
Figma's own research from 2025 found that 33 percent of designers are already using AI to generate design assets, and 22 percent are using AI to create first drafts of interfaces. The shift is not coming. It is already here.
What we do not know:
Whether speed of creation translates to quality of outcomes. Whether removing the design layer removes useful friction along with the unnecessary friction. How to maintain consistent visual quality and brand standards across a product that is built primarily through AI prompts. And whether the teams who move fastest are building things that users actually find better, or just things that ship sooner.
AI-generated code also carries meaningful risks. Research indicates it contains significantly more security vulnerabilities than human-written code, and Gartner has projected that by 2028, 25 percent of enterprise data breaches will trace back to AI agent abuse. Speed without oversight is not a benefit. It is a liability.
What This Means for Designers and Product People
The role that faces the most pressure in this shift is not the senior designer or the architect-level engineer. It is the mid-level production role, the person whose primary skill is executing designs in Figma or translating designs into code. That work is being compressed by AI tools faster than most organizations are prepared to admit.
The role that becomes more valuable is the person who can direct AI tools with precision, identify when the output is wrong and why, and bring judgment to a process that AI cannot replace. The skill is no longer knowing which button to press in a design tool. It is knowing what good looks like and being able to communicate that clearly enough for an AI to act on it.
This is not a small shift in responsibilities. It is a fundamentally different definition of what a designer or product person does.
My Own Experiment: What I Am About to Find Out
I have been designing digital products since 2006. For most of that time, some version of a dedicated design tool has been central to how I think and how I work. The mockup, the prototype, the handoff, these are not just steps in a process for me. They are how I discover what I actually think about a problem.
But I want to be clear about something. I did not arrive at this experiment overnight, and I am not suggesting anyone throw out their existing tools without context. My own workflow has evolved in stages, and understanding that progression is important.
A couple of years ago, I started incorporating AI features directly inside Figma. Figma's own built-in AI capabilities made it faster to generate design assets, explore layout variations, and produce first drafts of screens without starting from a blank canvas every time. That felt like a genuine improvement, not a replacement of the thinking process but an acceleration of the execution part.
Around the same time, dedicated AI design tools started entering the picture. Tools like UXPilot, which is built specifically to generate UI screen designs from a text description, made it possible to explore multiple directions quickly before committing to refinement in Figma. These tools do not replace design judgment but they do compress the time between having an idea and seeing it on screen.
For the frontend code layer, I also started using Claude AI to generate working HTML and CSS directly from design descriptions. Instead of a developer spending days interpreting a Figma file, a working frontend could be produced in hours from a clear prompt. That collapsed the traditional handoff step considerably.
Each of those additions felt like a meaningful step forward. And each one raised the same underlying question: if the tool can handle more of the execution, what exactly is the designer's job now?
Now I want to take the final step in that progression. Instead of starting in Figma and augmenting it with AI at various points in the process, I want to remove the Figma starting point entirely. For the next few months, I plan to build complete, working prototypes directly using vibe coding tools, specifically Replit, and see what that does to the quality, the speed, and honestly, the thinking.
I already have a sense of where the difficulty will show up. Keeping visual consistency and design quality intact across a product built entirely through prompts, without a structured design file to anchor decisions, feels like the hardest part. One of the challenges I am anticipating is getting consistent results when working against an existing codebase, maintaining the same visual language across different screens without a design system file to refer back to.
But I want to find out rather than assume.
I will write about what I discover here, not when I have clean answers, but while I am in the middle of finding them. The messy middle of an experiment is usually where the most honest observations happen.
So What Does Happen When Designers Start Prompting?
Honestly, we are still finding out. And that is the most truthful answer anyone can give right now.
What we do know is that the tools are capable enough to make the question real. This is not a hypothetical shift happening somewhere in the future. Designers and product teams are making this choice today, and the results are uneven enough that confident declarations in either direction should be treated with suspicion.
The designers who seem to be navigating this well are not the ones who abandoned their judgment when they abandoned their design tools. They are the ones who carried their thinking forward into a new medium. The prompt became the design brief. The AI became the execution layer. The designer became the person who knows when the output is right and, more importantly, when it is not.
Whether that is better or worse than what came before depends entirely on what you bring to it.
I will keep writing about what I find as I find it. If you are running your own version of this experiment, I would genuinely like to hear what you are noticing.
Frequently Asked Questions
What is vibe coding in simple terms? Vibe coding means describing what you want to build in plain language and letting an AI tool build it for you. Instead of writing code or designing screens step by step, you direct the outcome and the AI handles the execution. The term was coined by Andrej Karpathy in February 2025.
Is Figma being replaced by AI tools? Not entirely, and not yet. Figma itself has launched AI-powered features to stay relevant. But the traditional workflow of designing in Figma and then handing off to developers is being compressed significantly, especially for early-stage products and prototypes. For large organizations with established design systems and brand standards, Figma remains central to how product teams operate.
Will AI tools replace designers? Senior designers who can direct AI tools, define quality standards, and make judgment calls the AI cannot make are becoming more valuable, not less. Mid-level production roles face the most pressure. The skills that matter most are shifting from tool execution to design judgment and clear communication of intent.
What are the risks of building products with AI tools? The main risks are consistency, security, and quality of thinking. AI-generated products tend to drift from established visual standards without careful oversight. AI-generated code carries higher security vulnerability rates than human-written code. And there is an open question about whether removing the deliberate design process also removes useful cognitive work that leads to better products.
Which AI tool should I start with? It depends on your context. Lovable is the most accessible starting point for non-developers. v0 is strong for interface-focused work. Replit is best for full products that need data and back-end logic. Bolt is the most versatile general-purpose option. Figma Make is the natural entry point if you already work in Figma.