Can AI agent(s) COMPLETELY replace architects and designers?
”AI replacing X/Y/Z” is becoming somewhat banal nowadays, but I can’t stop thinking about it. This article from Sequoia Capital really pushed me down this thought exercise.
Looking at coding agents like Cursor or Claude doing real engineering work reminds me of when I first learned how to code several years ago. I studied the basics, wrote Python scripts line by line, struggled to debug, got excited when things finally worked, and sometimes didn’t even know why they worked 🤣
I genuinely enjoyed learning things the hard way. But those days are gone.
Even a year ago, most engineers I knew were using AI as a copilot. Now, they say most of their code is written by agents and they act more like conductors or supervisor. It’s crazy to think what’s going to happen in the coming months/years.
So will this breakthrough ever happen in architectural design too?
Until recently, I was skeptical. LLMs don’t have architectural taste. They can’t think three-dimensionally. They can’t even read simple drawings. There is not enough data to train the LLMs. It’s just not there yet.
But watching how fast AI agents are evolving, I’m starting to question whether it’s naive to think architecture is somehow an exception. These models are becoming insanely good at all kinds of knowledge work.
Here’s the bold idea: The real disruption isn’t AI tools for architects. It’s AI replacing the entire architectural service.
What if a company owned a full suite of AI agents handling the entire architectural design service? Not just conceptual design, the WHOLE thing.
/proposal. /concept. /floor-plan. /site-design. /detailing. /code-check. /documentation. /visualize. /respond-rfi. /budget-check. So on and so forth.
Human employees collect real-world data to train AI models, create local contexts, review and correct the AI outputs, manage client relationships, stamp the drawings, and monitor the token spend.
The winner will be the one who (1) dominates the data (2) engineers the entire process into natural languages and (3) replaces the full service, not just parts of it. Their customers won’t be architects. They will compete with human architects.
And here’s the key: those who attempt to build this with rule-based automations will absolutely fail.
No two buildings are the same because no two sites are the same. Every project is basically an edge case. That’s why productization of architectural service has been so difficult. (I learned this hard way at WeWork and Cottage!) The winning approach won’t rely on preset rules. It will require AI agents that can think and reason like human architects to navigate ambiguity and project-specific constraints. That’s exactly where LLMs shine.
Do I think there is a massive opportunity and paradigm shift in this? Yes.
Am I really excited about this future? I’m on the border 🥴