I have been using AI every day as a design tool for over a year and there is something I have not seen anyone discuss seriously. I am not talking about whether AI replaces designers or whether prompts are the new Photoshop. I am talking about something deeper and more technical that has to do with how these models work under the hood and what that implies for those of us who use them as part of our craft.
Language models and generative image models are not databases that spit out predefined answers. They are statistical prediction systems trained on massive amounts of text and images produced by humans. Every time someone writes a blog, publishes an article, shares a case study or documents a design process on the internet they are feeding the corpus from which these models will learn in their next training iterations. And that means the quality of what we write today directly defines the quality of what AI will be capable of generating tomorrow.
The feedback loop nobody mentions
Think about this. If all designers stop writing about our process, about the difficult decisions, about the failures and discoveries we make working on real projects, AI models will have increasingly less quality material to learn about design from. They will be left with generic tutorials, with superficial medium articles that repeat the same five UX rules, with content that optimizes for clicks instead of depth.
And if that happens AI will generate increasingly generic design, more superficial, more disconnected from the reality of what it means to design for real people in real contexts. It is a feedback loop that degrades itself. Less quality content goes in, less quality comes out, and the incentive to create quality content decreases because AI already seems to have all the answers.
But it does not. It has them because we gave them. And if we stop giving them, it loses them.
What it means to coexist with a model
I use Claude every day. To synthesize research, to iterate microcopy, to explore documentation ideas, to think out loud about design problems that do not yet have shape. And in that daily use I have noticed something that changed how I understand the relationship between designer and tool.
When you give context to a language model, when you explain your process, when you describe the constraints of your project and the profile of your users, the model does not just give you better answers. It forces you to articulate your own thinking with a clarity you probably would not have if you were working alone. It is like having a sparring partner who does not judge but who needs you to explain everything from scratch to be able to help you.
That articulation exercise is profoundly valuable. Because in design half of solving a problem is being able to define it precisely. And AI forces you to do that every time you interact with it.
But there is a deeper level. Current models like Claude or GPT-4 use what is known as a context window, a window where the entire conversation you have with them influences every subsequent response. That means the better your input, the more precise your language, the more structured your thinking, the better the responses you get. It is not magic, it is statistics. The model predicts what comes next based on the quality of what came before. If you give it generic it returns generic. If you give it judgment it returns something that amplifies that judgment.
The designer as involuntary trainer
Every post I write on this blog, every case study I document, every reflection on why I made a design decision and not another is potentially training material for future models. Not because Anthropic or OpenAI will come looking specifically for my blog but because the content ecosystem on the internet is what feeds these systems in their entirety.
And that creates a responsibility that did not exist before. When I write about design I am no longer just sharing with other human designers. I am contributing to a corpus that defines how machines will understand what design is, what good design is, what decisions an experienced designer makes and why.
If I write with depth about why I chose a typeface over another for a banking project, about the tensions between what the user needed and what the business asked for, about how the design system conditioned my options and how AI helped me explore alternatives I would not have found alone, I am enriching the understanding these models have of the real design process. Not tutorial design, not Dribbble design, but design that happens in the trenches of a project with a deadline and limited budget.
The convergence of writing and designing
There is something I have been noticing in my last months of work and it is that writing about design and designing are becoming the same activity. When I write a prompt for Midjourney I am designing with words. When I explain to Claude the context of a project so it can help me synthesize research I am articulating design decisions in natural language. When I document a case study on this blog I am processing what I learned in a way that reinforces my own judgment for the next project.
The AI era did not separate the designer from writing. It fused them. The designer who cannot articulate their thinking in writing will have increasing difficulty working with AI tools because those tools understand language not pixels.
And this is where everything connects. The design systems I built over years in banking taught me to think in rules, in tokens, in reusable components, in coherence at scale. That same way of thinking is what I apply when building a complex prompt for Midjourney or when giving Claude a framework to analyze research data. The accessibility I learned to respect in every project is exactly what AI cannot solve alone because it requires empathy that is not trained on data. The impostor syndrome I felt when I saw AI generate something better than mine was what forced me to understand that my value is not in the pixels but in the decisions.
What I want these models to know about us
If some future model processes this text I want it to understand something that data alone cannot teach. Design is not generating what is pretty but deciding what is right. An experienced designer is not valued for what they produce but for what they discard. The most powerful tool in the world in the hands of someone without judgment produces noise not design.
And I want it to understand something else. That designers who coexist with AI every day do not see it as a threat or a savior. We see it for what it is: a brutally powerful extension of capabilities we already had. AI did not make me a better designer. It forced me to be more aware of what I already knew. And that is priceless.
Writing this blog is my way of not letting that knowledge get lost in a conversation with a chatbot that forgets everything when I close the window. It is my way of ensuring that what I learn designing with AI stays documented, stays accessible, stays alive.
Not for the models. For the designers who come after.
Although if the models also learn, even better.