In an era where AI Agents do everything for us, what form should design tools take? Over the past few months, I've been completely immersed in the mission of designing a new design canvas, closely monitoring the AI strategies of countless design editors, new interfaces, and emerging tools to answer this question.
The tools we use determine how we work and our productivity. That's why we decided to create an in-house design editor rather than an external design tool, optimized for how we work. The sense of responsibility I feel in setting the direction and boundaries of a product that our internal designers will face daily has never been heavier. Now, questioning a tool's reason for existence has become more than just a matter of convenience or efficiency—it's become a question about the existential justification for why this tool must exist in this era.
As AI advances, human intervention will approach zero, but paradoxically, at that point of 'zero', we must reconsider the justification for a tool's existence. Every morning as I design a new canvas, the question I face ultimately converges to one:
In an era where AI advances to the point where human intervention approaches zero, what form should design tools for humans take?
This question is no longer in the realm of functional specs or technical choices. It's a fundamental question about the very reason for existence of the tools we create.
The Role of Canvas in Design Process
To answer this question, we must first look back at what the canvas was in existing design processes.
The traditional design workflow roughly followed this structure:
Empathize → Define → Ideate → Prototype → Handoff → Implement
In this flow, the canvas wasn't simply a 'drawing space'. It was a medium for externalizing the designer's thinking. A cyclical process of bringing vague intentions from the mind into visible form, manipulating them, and judging them again. Tools changed from Photoshop to Sketch, from Sketch to Figma, but the essential role of the canvas remained the same: making thoughts tangible.
On the canvas, designers arranged frames, combined components, and adjusted spacing while refining their intentions. What's important here isn't just the outcome. Judgment is formed through the repetition of changing layouts, undoing, and trying again. The canvas was both a tool for producing results and a tool for training the designer's thinking.
But AI began compressing the intermediate steps of this process. Directly from research to visualization, directly from ideation to prototype. As the process compressed, the cycle of thinking that happened on the canvas—the feedback loop of humans 'drawing with hands, confirming with eyes, and judging with brain'—was being entirely omitted.
The disappearance of process means the disappearance of the basis for judgment.
There's a more fundamental change. Until now, the canvas was ultimately just a blueprint. No matter how precisely a designer created a mockup, there existed a gap of translation where it had to pass through a developer's hands to become an actual product. But now, through AI, we can easily touch actual working code—production-level results directly. The ability to dramatically reduce the design-to-dev handoff cost, which had been a chronic bottleneck in the design industry for decades, was clearly a powerful wow point and the decisive reason why many people couldn't help but be enthusiastic about AI-based design (or development) tools. We've now reached a critical point where the intermediate translation layer called canvas itself could become meaningless.
From Creator to System Designer
Until just a few years ago, designers started with a blank canvas. They controlled every pixel, stacked layers, and created something from nothing. That manual process itself was the designer's skill and identity.
Now it's different. When you convey intention to AI, 5-6 layouts are instantly generated. Initial drafts, copywriting, asset search, layer naming—manual tasks that designers used to spend countless hours on have completely moved into the realm of automation. The era of computational design that John Maeda spoke of has truly arrived.
Faced with this change, the designer's role isn't disappearing but being reorganized. Now designers must become system designers who design the logic and constraints that enable AI to produce optimal results, rather than people who draw directly. It's not about remaining as curators who simply choose one from AI's options, but about defining the system itself—the context in which this massive computational device called AI should operate.
But here arises a fundamental question. Do system designers need the existing canvas? If the canvas has been a tool for externalizing thought, now that the subject of generation has shifted to AI, what intentions of the designer should the new canvas externalize?
The Paradox of Efficiency: When Friction Disappears, So Does Meaning
The greatest value AI brings is efficiency. It reduces human cognitive burden, eliminates repetitive tasks, and quickly produces results. Isn't that good?
But in design, friction isn't simply inefficiency.
The composition discovered by chance while trying layouts multiple times. The sensation felt while making subtle color adjustments. The judgment built while discarding failed mockups. All of this stems from friction in the process. The value of design lies not only in the outcome but in the process of thinking that leads to that outcome.
When AI removes all friction, designers work efficiently but simultaneously lose the ability to assign meaning. As Luciano Floridi pointed out, modern AI isn't designed to replicate 'intelligence' but to achieve 'performance' without intelligence. We delegate decisions to machines to avoid the pain of thinking. What we gain is efficiency, but what we lose long-term is the very ability of autonomous decision-making.
As someone creating design tools, what I'm wary of isn't a future where designers disappear. If AI could produce perfect results without human intervention, that would mean the tool has perfectly fulfilled its destiny.
The truly frightening scenario isn't designers disappearing, but tools only providing 'correct answers' while we forget to ask the question 'why should we make this?' A state where technical efficiency remains in place of vanished human labor, and no one asks where that efficiency is headed. That's the true 'void' I don't want to face as a tool creator.
The Trap of Parametric Reductionism
The process of AI producing optimal design ultimately involves reducing human taste, culture, and aesthetic values to quantifiable parameters. The aesthetics of imperfection that are hard to standardize or accidental innovations are eliminated as statistical outliers.
I call this Parametric Reductionism.
AI image generators show excellent skill in technical composition and balance. But they can't capture what Henri Cartier-Bresson called the 'decisive moment'—that emotional immersion and narrative clarity. Because you can't create something that deviates from the average by recombining past data to derive average values.
Nevertheless, companies adopt AI results for cost reduction and speed. This results in all digital interfaces converging to data-driven average values, making design increasingly homogenized.
This is the point I keep pondering while designing the canvas. Will our tool be one that accelerates convergence to the average, or one that breaks through the average?
Convivial Tools as an Old Answer
In 1973, Ivan Illich warned that when industrial society's tools pass a certain threshold, they lose their original purpose and instead reduce humans to appendages of tools. He called this Radical Monopoly. Just as pedestrians are excluded in car-centered cities, when tools completely dominate society, participation itself becomes impossible without them.
Fifty years later, the wholesale adoption of AI design tools is creating a new form of radical monopoly. An environment where you must accept the optimal results derived by algorithms to survive in the speed competition. A reversed relationship where technology uses humans.
The three criteria of Tools for Conviviality that Illich proposed are surprisingly valid for designing design tools in the AI era.
- Autonomy: Tools should enable humans to function as active creators, not passive consumers.
- Empowerment: The operating principles of tools shouldn't be hidden in black boxes; humans should be able to understand and control that logic.
- Enjoyment: Beyond result-oriented efficiency, the joy of the process itself—making decisions and assigning meaning—should be preserved.
Applying these three to canvas design makes the direction a bit clearer.
Intentional Friction as a Design Principle
When all cognitive obstacles are smoothly removed, users become passive recipients. Conversely, intentionally leaving cognitive space where designers can deeply contemplate and intervene at moments of aesthetic and ethical decision—this is intentional friction.
This is where I think the core of AI-era design tools lies.
When AI generates 10 mockups, instead of saying "Option 3 is optimal," the tool should provide designers with the context to judge for themselves why it's Option 3. Present results but leave decisions to humans. Provide efficiency but don't replace thinking.
This doesn't simply mean adding confirmation procedures for responsibility purposes. It means beautifully designing cognitive intervention points—like visualizing the basis data for AI-suggested results or pre-simulating user experience changes that would occur when a specific layout is chosen. The interaction method between humans and machines should change according to the level of autonomy.
- Assistance Stage: AI performs limited tasks according to prompts, and humans control all decisions.
- Partial Autonomy Stage: AI presents multiple alternatives, and humans select strategic directions.
- Conditional Autonomy Stage: AI operates independently within constraints set by humans, but humans intervene in exceptional situations.
What's important is that as autonomy increases, the frequency of human intervention decreases, but the qualitative importance of intervention grows exponentially. Tools must reflect this paradox.
As Tools Give More Answers, Questions Become Human Weapons
Jaron Lanier said, "There is no such thing as AI." Large language models or generative AI aren't independent intelligence but "innovative forms of social collaboration" that statistically recombine data produced by countless humans.
From this perspective, it becomes important to open the black box of AI tools and reveal the traces of people embedded within. Tools should transparently reveal their limitations as recombinations of past data, and humans should make judgments that machines cannot reach.
Don Norman has long criticized educational systems where designers are only immersed in visual techniques without understanding the complex social interactions of behavioral science, technology, and business. Now that AI perfectly replaces technical techniques, deep understanding of human behavior, moral reasoning, and vision presentation become designers' core competitiveness.
What We Must Create
In summary, design tools for the AI era should take this form:
- Tools where machines generate results, but the decision-making power to assign 'value' to those results is returned to humans.
- Tools that maximize execution efficiency but don't replace critical thinking toward purpose.
- Tools that present statistical averages but help designers break through those averages.
As AI's speed in presenting correct answers increases, what we need isn't the skill to accept correct answers. It's the insight to question and define 'why should this exist?'
What I aim for while designing a new canvas is clear. I don't want this tool to become a shield for preserving specific job roles. Even if all production processes are automated and physical labor disappears, I must aim for a sophisticated interface that helps preserve the designer's intention—the essence of the product. Helping only the designer's vision remain most purely in place of vanished production constraints—that's why I'm creating this tool.
In place of the repetition of arranging, adjusting, and undoing, leaving only the 'subject of intention' who fiercely contemplates what is truly valuable.
That's the only reason why tools for humans must exist, even in an era where human intervention approaches zero due to AI.