Since 1993, FORUM D&P has been documenting its architectural journey. Discover how we bring emotion into space through Archi-Therapy. Explore projects where architecture, technology, and human experience converge.

Full width home advertisement

Post Page Advertisement [Top]

200 Lines of Architecture

Where code meets architecture — the essence of both is simpler than we think.

In February 2026, Andrej Karpathy published something remarkable: the entire GPT algorithm — training and inference — in 200 lines of pure Python. No dependencies. No frameworks. Just the essence.

"I cannot make this any shorter," he wrote.

As an architect who has spent three decades translating buildings into data — from hand-drawn plans in a Parisian atelier to BIM models spanning four university campuses — that sentence stopped me cold. Because I've been trying to do the same thing with architecture.

The Essence of a Building

What are the "200 lines" of architecture?

Strip away the construction documents, the engineering calculations, the material specifications, the regulatory compliance, the contractor coordination. Strip away everything that is scale and engineering. What remains?

A single question: How does this space change the person who inhabits it?

That's Archi-Therapy — the philosophy I founded FORUM D&P on in 2008 with my partner Michelle LEE. Architecture as healing. Space as medicine. Every project we take on, from a Parisian apartment renovation to a national university campus masterplan, begins with this irreducible question.

Karpathy showed us that GPT's intelligence emerges from a simple pattern: predict the next token. Everything else — the RLVR training, the massive datasets, the GPU clusters — is scale. The pattern is 200 lines.

Architecture works the same way. The intelligence of a building emerges from a simple pattern: respond to human need. Everything else — the BIM coordination, the structural analysis, the MEP systems — is scale. The pattern is one question.

The Magnitude 9 Earthquake in Architecture

In December 2025, Karpathy posted what he called a "Magnitude 9 Earthquake" confession:

November: 80% manual coding + 20% AI
December: 80% AI + 20% manual coding
In one month, the workflow completely flipped.

I experienced the same earthquake — not in coding, but in running an architecture firm.

FORUM D&P is a small practice with offices in Seoul, Paris, and the US. We design buildings, plan campuses, conduct spatial research. The kind of work that traditionally requires teams of people spending weeks on analysis, documentation, and coordination.

In early 2026, we began an experiment we call AX — Agent eXecution. Five AI agents, each responsible for a domain of our practice: leadership orchestration, communications, design execution, research, and operations.

The results have been Karpathy's "10X" — but in architecture.

On a single day in April 2026, our AX system processed 30 unread emails, prepared cost estimates for two university projects (one space survey, one campus masterplan across five campuses), drafted professional correspondence, created project folders with work logs, coordinated with team members via Microsoft Teams, and generated operational procedures — all while I focused on the decisions that actually matter: pricing strategy, client relationships, and the quality of our architectural thinking.

What used to take me an entire week now takes an afternoon of approvals.

Vibe Architecture

Karpathy coined "Vibe Coding" — the practice of giving AI a direction and letting it handle the implementation. A throwaway shower thought that became industry-standard terminology.

I'd like to propose its architectural counterpart: Vibe Architecture.

Vibe Architecture is what happens when an architect stops being a draftsperson and becomes, fully, a thinker. When the 80% of our time that used to go to documentation, coordination, and repetitive analysis is handled by AI systems — and the 20% that remains is pure architectural judgment.

This isn't about replacing architects. Karpathy's microgpt doesn't replace machine learning engineers. It reveals that the algorithm is simple — what's hard is knowing what to build and why.

Similarly, AI in architecture doesn't replace the architect's eye, the spatial intuition developed over decades, the understanding of how light falls through a window at 4 PM in November in Paris. What it replaces is the mechanical labor that kept us from spending more time at that window.

Data as Architectural Material

My career has been an extended experiment in treating data as an architectural material — as fundamental as concrete or steel.

In 2021, we began surveying Changwon National University's campus, translating buildings into spatial data. By 2025, we had completed masterplans for four campuses, analyzed 250,000 m² of building area, and created frameworks for facility management that universities across Korea now reference.

When Gyeongsang National University needed a similar space survey, we didn't start from scratch. We replicated our methodology, adjusted the parameters, and produced results in a fraction of the time. When Jeonbuk National University — five campuses, from Jeonju to Namwon — asked for a masterplan estimate, our AI analyzed 921 previous projects in our database, found the relevant benchmarks, and generated a calibrated cost estimate in minutes.

This is what Karpathy means by "10X." Not doing the same thing faster. Doing fundamentally different work because the mundane is automated.

The Question Remains

Karpathy's microgpt is a teaching tool. It doesn't build ChatGPT. But it reveals the mechanism — and in doing so, it frees us to think about what we're actually building and why.

FORUM D&P's AX experiment is the same kind of exercise. We're not trying to automate architecture. We're trying to reveal its mechanism — and free ourselves to focus on the question that has always mattered:

How does this space change the person who inhabits it?

The complexity comes from scale. The essence is always human.

200 lines of code. One question of architecture. The rest is engineering.


Inki LEE is an architect (DPLG, France) and CEO of FORUM D&P, a Seoul-Paris-US architecture and urban design practice. He is currently leading the FORUM AX project — an experiment in operating an architecture firm through AI agent systems. His work focuses on campus masterplanning, spatial data analysis, and what he calls Archi-Therapy: architecture as a medium for human wellbeing.

www.forumdnp.com | www.atat.cloud

No comments:

Post a Comment

Bottom Ad [Post Page]

| Designed by Colorlib