singularitypracticealmanac

I predict three modes of surviving practice: how architects have a second life.

Conventional practice follows the pyramid organization chart. Extended mind practice is one human thinking through a rig of tools. In decentered practice, humans step to the edge.

My own practice runs all three concurrently.

We are witnessing all meaningful work taken over by AI. The diagram places humans in relation to each field and the speed of its decision-making loop.

In finance, decisions happen in microseconds. Humans are already out of the loop. In self-driving cars, humans occupy a fallback position. Architecture is less neat: its modes of practice coexist.

The attended AI loop is how most of us have worked with AI so far: one human, one model, a serial conversation. Chatting with ChatGPT or Claude. Queuing a prompt in Midjourney.

In the agentic loop, the architect briefs an agent that can maintain a goal and plan its work. The human waits at a checkpoint while the AI works through the task.

This presentation was made this way.

The top-down hierarchy survives. Humans remain connected to every decision-making process. Peter Zumthor is not going out of business because AI is here.

Architects will continue to work in different modes.

This is where I am primarily occupied today. The tools we learned before AI form the inner layer. Generative AI—Midjourney, Magnific and others—forms another. Agentic AI works around them and can use those tools in its own workflows.

One human thinks through a rig of tools. I borrow “extended mind” from Andy Clark and David Chalmers, whose work Neil Leach opened my mind to.

The next displacement moves humans further from the work itself. Agentic AI takes on most coordination and production. The human moves to the side.

I predict that decentered practice will coexist with the other modes. My own work already runs all three concurrently.

The Center for Autonomous Systems and Technologies was completed in 2017 through a design-build project led by MATT Construction and CO Architects. I worked as designer and project architect from schematic design through punch list.

An existing courtyard became a covered arena for advanced robotics and drone testing. I designed the geometry of its ETFE tensile roof.

This was my first encounter with machine learning and robotics as the subject of an architectural project.

AI Sapien: Variations on Architecture and the Future sets dialogues with three AI systems alongside images made with Midjourney. Wes Jones wrote the foreword.

The dialogues ask AI about simulation and consciousness. The book is a creative work, not a manual. It received first place at the A+D Museum Awards and is held in more than sixty libraries, including the Getty and Princeton.

These machine architectures are portraits of a species of being I call AI Sapien. They developed from lecture slides for a SCI-Arc symposium in March and appear in Chimera, guest-edited by Elena Manferdini and Damjan Jovanović.

The name plays on Homo sapiens. Removing the final “s” leaves a dependence on humans. These giant machines may be more capable than us, but I imagine a hopeful relationship of symbiotic cooperation.

The Busan Museum of Art commissioned this work for an exhibition in 2026. The exhibition places a digitally imagined museum within the public experience of the city.

I work in steps. The initial image begins with seed images and mood boards, with the rendering settings deliberately lowered.

My working premise is that a generation has a finite processing capacity: asking for novel form and photorealism at the same time can muddle both. I resolve the form first, then build texture and detail in successive layers.

The curator’s note describes an increasingly intimate convergence of virtual and physical realities: a future museum shaped through AI and digital technologies, made available as a sensory experience in public space.

These images need not remain two-dimensional. Image-to-3D tools can translate an image or sketch into an object that a designer can orbit, inspect and develop. This example uses Tripo and is shown in Rhino.

The model opens another stage of work: an image becomes geometry that can be examined from multiple directions.

Gemini rendered the aerial and wireframe views. The museum sits within a simulation of a virtual Busan.

The third case study is this presentation. Claude drew its diagrams through an agentic loop with sub-agents.

The starting point is loose and messy: my parti sketches, drawn on paper and photographed. For me, idea generation still begins with a pen.

I translate the sketch into a written brief and seed the conversation with my own diagrams. Plain language establishes the criteria for what follows.

The agent can read my markups. I can ask it to align a legend, revise a relationship, or change the organization of a diagram. The output comes back for judgment.

The form develops through repeated correction. Each iteration needs to become more elegant and more sensitive to the subject.

Why keep a human in the loop when superintelligence arrives? My answer is deliberately cynical: someone needs to be fired or sued.

That someone is the human. My argument is that architects still depend on their own personhood. Responsibility goes further than curation.

Project these practices forward. The discipline divides like a cell, with more computational power and more agents available to an operator.

Practices split, spawn and multiply. Eventually, we may no longer be sure whether the people within them are humans, machines or something else.

In my book, I call these systems “thing-beings.” They occupy a liminal condition. Their contributions are named here.