AI-generated editorial visualization · not a real client eventA physical project is rude in a useful way. It refuses to be persuaded by confident software language.
A dock or outdoor-space concept has to survive use, access, weather, materials, code, permitting, engineering, budget, sequence, and taste. Miss one hard constraint and a week of handsome work can become scrap.
I start before the rendering
“Design a dock” sounds like a task. It is really a bundle of decisions waiting to be ordered. What must the space enable? Which constraints are fixed? What needs a site visit? Which answer changes every answer after it?
That is work an agent can help structure. I separate use cases from constraint research, develop materially different options, map dependencies, then challenge each option under changed assumptions. The lanes converge into a comparison, not a beauty contest.
A rendering can help someone react. It cannot prove that a concept is feasible, permitted, safe, or buildable.
The useful deliverable has some friction
I want a requirements brief, an assumptions register, a site-verification checklist, option tradeoffs, questions for engineers and officials, and the order in which decisions should be made. It is less glamorous than a perfect image and far more useful in the next meeting.
The boundary is load-bearing
Agents can organize research, compare alternatives, and catch contradictions. They cannot inspect a site, certify structural adequacy, issue a permit, or accept construction risk. Licensed professionals and relevant authorities must confirm the matters in their jurisdiction. Contractors own their means and methods. The owner owns the tradeoffs.
My standard is not “replace the professional.” It is “arrive prepared enough to use the professional’s time well.” Better questions, earlier conflict discovery, and a clean decision record are real gains. Pretending the plan has already touched the ground is not.
← Back to Blog