My Partner Has No Name Yet

Ibrahim Fawakherji — ArchUp

In the first months of 2007, a senior executive at Microsoft was asked in a press conference what he thought of the newly announced iPhone. He laughed. Not a polite laugh, but the particular laugh of a person who has seen enough technology cycles to feel entitled to certainty. He asked, rhetorically, who would want a phone without a keyboard. The question was meant to end the discussion. It did not end the discussion. It ended his credibility on the subject, and he has been living inside that laugh ever since, a footnote to a moment he was certain he understood.

I tell this story not to mock him. I tell it because I bought an iPhone in the first wave and returned to my Nokia within two weeks. The early device was incomplete, inconsistent, genuinely difficult to use for the things I needed to do. I was not wrong about the specific product. I was wrong about the direction. By 2011 I had come back and have not left. The technology had not changed my assessment. Time had.

I have been writing about AI and architectural practice for long enough that I am now accountable to my own earlier positions, which is a condition I prefer to the alternative. I said, in those earlier pieces, that we were in a period of crystallization, that the tools were forming and reforming faster than expertise could accumulate around any of them, and that the correct response was patience and selective engagement rather than either wholesale adoption or principled refusal. I still hold that position. But something has shifted in the landscape, and the shift is large enough that it requires a new piece rather than an amendment to an old one.

The era of the hardware race is effectively over.

What is beginning now is the era of the agent.

The smartphone wars of the 2010s followed a pattern that the AI industry is now replicating with striking fidelity. Two primary competitors, iPhone and Samsung, achieved a kind of technical parity that made the marginal improvements of each new release feel increasingly incremental. Camera resolution, battery life, processing speed: each cycle produced genuine advances, but the advances were narrowing toward a ceiling that the underlying physics of the hardware was approaching from below. The competition shifted from capability to refinement, from breakthrough to optimization.

The AI hardware race has followed the same arc. The foundational models that emerged between 2022 and 2025 achieved a level of capability that subsequent releases have been iterating around rather than transcending. The competition between platforms, GPT against Gemini against Claude against the next arrival, has produced improvements that matter at the margin but do not fundamentally change what the technology can and cannot do. The practitioner who moved between platforms every few months in search of the one that would finally transform their workflow was chasing the same marginal differences that smartphone buyers chased when they switched between flagships every upgrade cycle.

What changes the equation is not a better model. It is a different relationship with the model.

I want to record this position against something that has nothing to do with AI, precisely so that neither I nor the reader can move the date later to suit the argument. Today is Friday, August 7, 2026, and at the time of writing, gold is trading at approximately $4,345 per ounce. I am placing the price here deliberately as a timestamp. If you return to this article five, ten, or twenty years from now, when agents may have become something entirely different, come back to this paragraph. If I was wrong, tell me: Fawakherji, on Friday, August 7, 2026, when gold was around $4,345 an ounce, you said the agent would become the next major relationship between the architect and AI, and you were wrong. I will accept the correction. And if, by then, architects are working through persistent agents that know their projects, files, habits, correspondence and objectives, then this paragraph will have served its other purpose: fixing the argument to the conditions in which it was actually made, before the outcome became obvious. I will record one more position beside it: on that same Friday, with gold at $4,345 an ounce, I still do not believe AI can replace the architect who possesses a genuine architectural philosophy. If that changes too, come back and correct me.

The agent is that change.

An AI agent, in the current technical sense, is not a tool you query. It is a system you configure to operate on your behalf within defined parameters, executing sequences of tasks, managing communications, processing information, and maintaining context across time without requiring your direct involvement in each step. The distinction between querying a model and deploying an agent is the distinction between asking someone a question and giving someone a brief. The first transaction ends when the answer arrives. The second transaction continues until the objective is achieved or the parameters require revision.

For architectural practice, the implications are specific and significant. The administrative load that consumes a disproportionate share of a practitioner’s cognitive bandwidth, client communications, document management, scheduling, research aggregation, permit tracking, contractor correspondence, the entire infrastructure of a project that surrounds and supports the design work but is not itself design, this is the domain where agents will produce the most immediate and measurable return. The time recovered is not incidental. It is the time that would otherwise have gone to tasks that require presence but not judgment, attention but not expertise. Returning that time to the practitioner changes what the practitioner can do with a day.

The more important question is what they are becoming.

And today, Friday, August 7, 2026, my answer is this:

They are becoming partners.

Not partners in the philosophical act of architecture.

Not yet.

But partners in protecting the architect from the enormous machinery surrounding that act.

The architect who spends the next few years memorizing every prompt on every platform may discover that the entire interface has disappeared beneath him.

The architect who spends those same years developing judgment, philosophy and a recognizable intellectual position, while learning how to delegate effectively to intelligent systems, may find himself with something considerably more valuable.

Time.

Time to look.

Time to question.

Time to draw.

Time to reject the first solution.

Time to think.

And perhaps this is why the title of this article remains deliberately unresolved.

My partner has no name yet.

Today it may be one model.

Tomorrow another.

But I want to be precise about the boundary, because the boundary is where the important work lives.

I stood in the Fondation Louis Vuitton in Paris and looked at Frank Gehry’s preliminary sketches for the building. Not the finished drawings. The first marks, the quick lines that record a mind in the act of proposing something to itself before the proposal has a fixed form. I have written about this elsewhere in this series. What struck me then, and strikes me more forcefully now in the context of this argument, is the question of what those sketches actually contain.

They contain a philosophy. A set of accumulated convictions about what architecture should do to a person standing inside it, about the relationship between structural fact and spatial experience, about what materials owe to the light they inhabit. These convictions were not present in the sketches as explicit content. They were present as the grammar that produced the marks. You cannot extract the philosophy from the sketch and put it somewhere else. The sketch is the philosophy made visible, and the philosophy is the product of a specific life of looking and building and failing and revising.

I do not believe, and I am recording this belief with the same directness I applied to the agent question above, that AI will displace the architect who possesses this kind of formed philosophical position. The architect whose work is recognizable not because of a visual signature but because of a consistent way of posing the problem, whose buildings ask questions that other buildings do not think to ask: this person is not competing with a model. They are doing something the model is not designed to do.

What AI will displace, is already displacing, is the practitioner whose primary competitive advantage is access to technical tools, whose position in the market depends on the ability to produce outputs that are now available without the same human labor investment. This is the equivalent of the photographer who competed on darkroom technique in a world that moved to digital processing. The technique was real and the skill was genuine, but the technique was not the irreplaceable thing. The irreplaceable thing was the eye, and the eye survived the transition.

The practitioner preparing for the agent era needs to make two parallel adjustments, and they move in opposite directions.

The first is to delegate more aggressively than feels comfortable. The tasks that agents will handle competently are tasks that currently consume time and attention that could be directed elsewhere. Holding onto those tasks because they feel familiar, or because relinquishing them requires an upfront investment in configuration and trust-building with a system, is the equivalent of continuing to hand-calculate structural loads after the software arrived. The calculation was meaningful work. It was also no longer the best use of the available cognitive resource.

The second is to invest more seriously in the thing that cannot be delegated. The design philosophy, the analytical framework, the accumulated capacity to read a site and understand what it is asking for, the ability to hold a client’s actual need in view while navigating the gap between that need and what the client thinks they want: none of this is producible by a configured agent. It is produced by practice and by the particular kind of attention that sustained engagement with difficult problems develops over time.

The agent handles the administration of the practice. The architect remains responsible for the intelligence of the practice.

These are not competing claims. They are a division of labor that the technology is making possible for the first time at a cost that small and medium practices can actually access. The senior partner at a large firm has always had administrative infrastructure. The independent practitioner has always been their own administrative infrastructure. The agent does not change what architecture requires of its practitioners. It changes who can afford to focus exclusively on what architecture requires.

The iPhone executive laughed at a device that was genuinely incomplete in January 2007. He was not wrong about the specific product in front of him. He was wrong about what the product was becoming, and he was wrong in a way that foreclosed his ability to participate in the becoming.

The agents arriving in 2026 are incomplete. They will make errors that feel significant. They will require more configuration than the promise suggests and more patience than the demonstration implies. These are the growing pains of a technology in its iPhone-2007 moment, real and temporary in roughly equal measure.

What they are becoming is a permanent restructuring of the relationship between the practitioner and the practice. The architect who treats the agent as a threat to their professional identity will have the same relationship to this transition that the Nokia executive had to the touchscreen. The architect who treats it as a new kind of junior partner, capable and limited in specific and learnable ways, will find that the transition has returned something the profession has been losing for decades: the time to think.

That time is where architecture lives.

The agent’s job is to protect it.

✦ ArchUp Editorial Insight

The article’s central argument — that the agent era represents a qualitative shift from tool to partner — is structurally honest in its ambition but requires a more precise accounting of where the boundary between delegation and dependency actually falls, because the same analysis this archive applied in The Dematerialization of the Studio and Architectural Design for Twenty Dollars applies here with additional force: an agent configured to manage client communications, document workflows, and permit tracking is not a junior partner in any neutral sense — it is a subscription infrastructure that, once embedded in a practice’s operational nervous system, produces exactly the switching-cost dependency that platform vendors require to sustain pricing power, and the practitioner who has delegated aggressively enough that the agent is handling the administrative connective tissue of the practice has simultaneously made themselves structurally incapable of operating without it. The iPhone analogy the article deploys is accurate about the direction of travel and evasive about the terms: the person who returned to Nokia within two weeks was exercising a reversibility that the deeply agent-integrated practice will not possess, because the Nokia alternative was always there, whereas the administrative infrastructure that the agent replaced — the assistant, the office manager, the junior staff whose salaries the subscription now undercuts — will not be recoverable at the moment the platform reprices, changes its terms, or decides that the architectural sector is too small to prioritize when a larger market demands its attention, a condition that connects precisely to what Who Pays for the AI Hallucination established: the party who absorbs the cost of the system’s failure is always the party who integrated most completely, and the time the agent returns to the architect for thinking and drawing is real — but it is rented time, held at terms the architect did not set and cannot unilaterally revise.

Ibrahim Fawakherji — ArchUp