The Practice That Forgets Itself
Ask most principals what their practice's design language is and they'll tell you.
Ask them to write it down and it gets harder. Ask them to write it down precisely enough that someone else could use it without asking them? Almost nobody can do that.
Where the knowledge lives
The knowledge is real. It's just not written down anywhere.
It lives in the work. In fifteen years of projects that express the same underlying sensibility without ever making it explicit. In the instinctive choices made thousands of times — this material not that one, this proportion not that one, this quality of threshold not that one — that add up to a recognisable voice. A practice DNA that exists fully in the principal's head and partially in the portfolio and nowhere else.
This is fine when the practice is small. The principal is in every project. Their judgment is everywhere. The DNA travels with them.
How it breaks down
It starts to break down as the practice grows. New team members need to learn the language. Clients have to be briefed on it. Every new project requires re-establishing what the practice is, because there's no system that holds it. The principal briefs the team. The team produces work. The principal corrects it. The team revises. This is how it works. It's how it's always worked.
It's also slow. And it means the practice's quality is permanently capped by how much of the principal's attention it can access.
The deeper problem
The deeper problem is what happens over time. Practices forget themselves. The early work that defined the voice gets buried under the volume of later work. The instincts that made the practice distinctive get softened through the accumulated compromises of working with more clients, more briefs, more varied constraints. The DNA doesn't disappear. But it stops being actively referenced. It stops actively shaping the work.
New AI tools sit on top of this problem and don't solve it. They generate output — fast, professional-looking, technically competent. But they don't know the practice. They produce the average. Every project starts from zero, from the model's defaults, from architecture-in-general.
Extraction, not prompting
The answer isn't more prompting. It's extraction. Pull the DNA out of the portfolio, out of the reference images, out of the materials library, out of fifteen years of built work. Put it somewhere the tool can hold it. Then every output starts from the practice's actual language, not from a statistical average.
The practice stops forgetting itself because the memory is finally stored somewhere other than a person.
That's not a technology upgrade. That's the difference between a practice that sounds like something and one that's perpetually starting over.
Dave Mattu is the founder of Skalm — AI for architecture practices that holds your style, your voice, and your way of working.