singularity practice almanac

If prediction is the basis of intelligence, then this almanac is a prediction on new modes of working.

In the 30 August 2026 DigitalFUTURES broadcast, Neil Leach spoke about the death of architecture in light of rapid advances toward artificial superintelligence.

In response to Leach, on my broadcast on the same channel, I predicted three modes of surviving practice: how architects have a second life.

They are: conventional practice, showing the pyramid org chart; extended mind practice, one human thinking through a rig of tools; and decentered practice, where the human steps to the edge.

My own practice runs all three concurrently.

We are witnessing all meaningful work taken over by AI.

Shown here is the human relation to each field: the closer the human is to the loop, the decision-making process, the closer the human involvement. The diagram is organized by the speed to complete the loop.

In finance, AI takes everything in microseconds; humans are already out of the loop. Self-driving cars have humans at the fallback position, and so on. Architecture is not so neat, but AI is taking over everything.

The attended loop is how most architects and professionals are used to interacting with AI: one human, one model, in a serial conversation or prompting. This is like chatting with ChatGPT or Claude in a chat interface, or queuing up Midjourney.

The agentic AI loop is the forward way of interacting with AI. The architect, the operator, briefs the agent; the AI has become so capable that it now holds goals and can plan on its own. The human waits at the checkpoint while the AI works on its own.

How does the AI takeover, and the new ways of interacting with AI such as agentic, affect our practices? I argue three modes will survive, leaner and stronger.

The first is conventional practice. This is the org chart we are all accustomed to: a top-down hierarchy with a human attached to every position. The current default.

Peter Zumthor is not going out of business because AI is here. Architects will still have various modes of operation, is what I predict.

This is where I am primarily occupied these days. A new mode of practice has emerged in which one human is amplified by conventional and novel means of AI collaboration. This is the extended mind practice, based on Andy Clark and David Chalmers's argument that the mind extends into its tools, which Neil Leach opened my mind to.

The inner layer is conventional tools, the middle layer generative AI, the outer layer agentic AI. The agentic AI can use the tools of the inner layers as sub-agents.

The way forward will see a further displacement of humans from the work itself. In decentered practice, agentic AI and its sub-agents can run an entire practice and organization, including most design work. Agentic AI takes care of most coordination and work; the human steps to the edge.

This mode of working is going to exist along with the other modes, is what I predict. This presentation was made in part using this method; the steps follow in the third case study.

Caltech CAST was completed in 2017 as a design-build project led by MATT Construction and CO Architects. I was involved from schematic to punch list as the designer and project architect, while working at CO.

It is an adaptive reuse project that converted a courtyard into a drone testing arena with a covered roof. I designed the ETFE tensile roof geometry and the arena.

This project was my first encounter with machine learning and robotics. It is an example of conventional practice, in which AI was then the client and subject matter.

Case study two: three recent projects made with layers of AI, in the extended mind practice that I currently run.

This is my 2024 book, AI SAPIEN: Variations on Architecture and the Future (ORO Editions), foreword by Wes Jones. The book contains dialogues with three AI systems, set to images made with Midjourney.

Platonic dialogues ask the AI themselves on topics of simulation and consciousness; the book is a rare creative work rather than a manual. AI SAPIEN won first place at the A+D Museum Awards in 2023 and is in sixty-plus institutional libraries, including the Getty and Princeton University.

These are based on lecture slides for a symposium at SCI-Arc in March, and for the Chimera issue of AD, guest edited by Elena Manferdini and Damjan Jovanović. The machinic architectures are portraits: mechanical dwellings, as sentient machine entities.

AI SAPIEN is the term described for the sentient and sapient machines: AI and robotics. The term SAPIEN is a Derrida-inspired play on words based on Homo sapiens, in which the final "s" is removed in a purposeful misspelling to indicate absence and deferral of meaning. These beings, like us, are incomplete without symbiosis.

The final example of extended mind practice is the Simulacra Museum. The Busan Museum of Art commissioned the piece for an exhibition that ran in 2026. Here are the process and the steps.

I work in layers and steps. The initial image is generated from seed images and mood boards, with the graphic rendering settings purposely lowered.

Midjourney, as with most frontier AI, has a very finite processing capability. If you want photorealism with novel form, it muddles in the middle. So form first, then texturing. The realized materiality comes in layers, to additively combine the finite computational allotment of each prompt.

Diffusion-based images can be exported as three-dimensional output. Tripo exports to Rhino as geometry, in formats including FBX; the designer can orbit the image, model over it and translate it. Various competing models for direct-to-BIM are also emerging in the market.

The aerial and the wireframe, rendered by Gemini, have the hill-sited museum overlooking a virtual Busan.

For the third and final case study, of decentered practice, the case study is this presentation you are watching right now. These diagrams were drawn by Claude agents, running the agentic loops with sub-agents in tow.

These are the parti sketches, done on paper and photographed by me as documentation. It is loose and messy; for me, idea generation still starts with pen on paper.

Then the brief, as a conversation with the agent. I describe my work model and the goal to the agent in plain language, seed it with my own diagrams as reference images, and allow the AI to make its own decisions. The plain language sets the criteria after that.

Then I manage by markup. The agent proposes. I redline. The AI corrects.

The agentic workflow can read markups now. On the upper right, I told Claude to align the legend to the right, and it did.

In an iterative sequence the form of the diagram is improved and made more elegant and topic-sensitive. Two initial proposals are rejected here, and one is kept. This presentation is full of such micro and macro decisions by the operator and the agents. The human operator who is more critical, able not just to curate but to guide the direction, will be successful.

Why even have a human, when superintelligence arrives? I posit that someone needs to be fired or sued. That someone is you. As long as AI does not have legal personhood, we leverage human personhood in addition to drafting skills, in the singularity.

Architects depend on our own personhood, is my argument. It is a step more cynical than curation itself.

Let's project forward and run a simulation of what practice would be, from conventional practice to decentered. What we call architectural mitosis is when humans are able to field countless agents and artificial superintelligence.

The decentered discipline divides like a cell, showing the doubling of computational power available to the operator. Earth and the human remain in their central orbit at first, as superintelligence keeps increasing human potential and our civilizational reach. Practices split, spawn and multiply, until we are no longer sure if the people in them are human, or machines, or something else.

Are we human still?

Read more on AI SAPIEN: Variations on Architecture and the Future (ORO Editions, 2024)

Read more on Chimera: The Architecture of Contemporary Utopias (Architectural Design, Vol. 96.2, 2026)
Guest-edited by Elena Manferdini and Damjan Jovanovic