Reinier de Graaf's Manifesto - "Rely on AI for matters of taste"

I want to tackle De Graaf’s sixth point from his manifesto: ¹

6 Rely on AI for matters of taste. We design buildings – or so we think. More commonly we broker arbitrary choices of finishes, fittings, textures, colour palettes, or (worse) moods. Embrace AI – not to think for us, but to help us think more clearly, and rid architecture of the burden of frivolous decisions.

He is not calling for the automation of design. He is calling out a professional assumption, that the aesthetic decisions architects make day to day represent the exercise of cultivated expert judgment, that taste is a form of disciplinary knowledge, and that only a trained architect can navigate it responsibly. His claim is that most of what passes for aesthetic expertise in practice is something far more mundane: preference arbitration. Choosing finish A over finish B. Colour palette X over Y. Mood board this over that. Not because one is demonstrably better, but because someone in the room preferred it, or because it matched the brand, or because it felt right today.

I think he is right, but I think the argument goes further than he takes it.

Design decisions, including aesthetic ones, require justification. That justification can take several forms: direct empirical testing, established heuristics derived from cognitive and perceptual research, post-occupancy data, or environmental psychology findings. What it cannot take is the form of unchallenged personal taste, however cultivated. The burden is on the designer to say: here is what I am optimising for, and here is the evidence that my decision gets closer to it. If they cannot do that, “I am a trained designer” is not a sufficient answer.

We have seen what happens when a design field is forced to confront this. What was once simply called web design, and later rebranded as UI/UX, gradually developed data-driven methods to test whether aesthetic decisions actually produced the user behaviour they were assumed to produce. Marissa Mayer, Google’s VP of Search Products, made the case directly: “When I do user-interface design, the designer would come to me and would say there’s this green or that green. And we don’t need to make an arbitrary decision because we’ll just run both of them on the site, watch the data and the metrics that come out of that and we’ll be able to scientifically and mathematically prove which one the users actually seem to be responding to better.”

The most notorious example of this logic applied at scale was Google’s “41 shades of blue” experiment. Unable to decide between two shades of blue for its hyperlinks, Google tested every shade in between, serving each variant to a slice of users and measuring clicks. The winning shade generated an additional $200 million in annual revenue. Expert intuition was frequently wrong. Google’s Visual Design Lead, Douglas Bowman, quit the company in direct response, writing that the data-driven culture had reduced every decision to “a simple logic problem,” preventing the company from ever making a daring design decision. He had been asked to justify, with data, whether a border should be 3, 4, or 5 pixels wide. ^{2 3 4}

Bowman’s frustration is understandable, but his conclusion is wrong. The pixel border question is not unanswerable. It is answerable by reference to an existing evidence base about spacing, visual hierarchy, and perceptual grouping, much of which has been formalised into usability heuristics derived from decades of empirical research. ⁵ When interface elements follow a predictable pattern, users do not have to re-learn interactions. Error rates drop. Task completion times improve. A designer who chooses a typeface size based on legibility research and accessibility standards is making an evidence-based decision. A designer who chooses it because it feels right is not, regardless of how many years of practice stand behind that feeling. The correct response to a bad measurement instrument is a better one, not the abolition of measurement.

Architecture has not done this. It does not systematically test whether the material palette chosen for a residential development produces different resident satisfaction than the one that was rejected. It does not connect specific finish decisions to measurable behavioural or environmental outcomes. It makes an arbitrary aesthetic choice, moves on, and files the decision under expertise. The fee is charged accordingly.

This matters more than it might seem. Environmental psychology research, particularly in healthcare settings, shows that material choices such as texture, colour temperature, and surface warmth produce measurable differences in recovery rates, staff stress, and patient perception of safety. In housing, the materials used to clad a building send social signals that affect community pride and resident attachment to place, even when the floor plans are identical. If material decisions genuinely affect human outcomes in measurable ways, then leaving them to mood boards and client preference sessions is not expertise. It is negligence dressed up as expertise. The case for evidence-based decision making becomes stronger, not weaker.

Here is the uncomfortable implication. A developer with a large enough global portfolio, collecting post-occupancy data across thousands of buildings, could train an AI to make those decisions better than any individual architect working from intuition and a finish sample library. If that happens, the architect’s claim to expertise in design and aesthetics does not get automated away. It gets taken away, because architects never demonstrated it was expertise in the first place. Clients and developers do not wait for a profession to reform itself. They route around it. What remains for the architect is a technical and regulatory function: coordinating consultants, managing the building process, satisfying code. Valuable, but not what the profession has historically claimed to be, and not something that commands the same authority, fees, or trust.

There is also a behavioural dimension worth naming. Architects spend finite cognitive bandwidth on low-stakes aesthetic choices, crowding out attention for the decisions that actually require expert judgment: spatial organisation, programme, structural logic, relationship to context, climate response. Decision fatigue is well-documented. An architect who has spent three hours resolving a client’s preferences about tile finishes is not in the same cognitive state as one who has not. De Graaf’s proposal is straightforward: let AI handle the decisions that are testable or resolvable by established heuristics, and reserve human judgment for the ones where it genuinely cannot be replaced.

The resistance to this within the profession is worth examining. If taste is how architects understand their own identity, the thing that makes them artists rather than technicians, then delegating taste decisions to an algorithm feels like an identity threat rather than an efficiency gain. But that reaction is precisely the one that should make us suspicious of our own reasoning. The honest question is not whether aesthetic decisions matter. Some clearly do, and the evidence base to support them exists or can be built. The honest question is whether the current process of making them, by assertion of expertise, by client preference loop, by intuition defended as knowledge, is getting us closer to a practice where architects actually control design decisions and can defend them. I think it is not.

But freeing up cognitive bandwidth and defending design decisions with evidence only solves half the problem. Even if architects succeed at both, the capital problem remains. De Graaf identifies it plainly: architecture has become a tool of capital. Developers do not hire architects because they value design. They hire them because architectural involvement helps secure planning approvals and provides cultural legitimacy for financial extraction. As he writes in The Baffler, “too often architecture serves as a fig leaf for financial returns.” ⁶

His solution is political: collective organisation, direct user relationships, independent agenda. I think this is necessary but not sufficient. Because even if architects build collective authority and refuse to serve developers, whoever finances construction will still shape the brief. Removing the developer from the equation does not remove the dependency. It just leaves a gap that someone else will fill.

The more uncomfortable proposition is that architects need to evolve their practice to include the financing of construction. Not as a concession to developer logic, but as a way of escaping it. The reason architecture currently serves as a fig leaf for financial returns is structural: the people who hold capital and the people who hold design knowledge are kept apart. The developer captures the surplus because they control the financing. That will not change by architects becoming better at politics. It changes when architects become capable of holding both.

The objection that mixing design and financial incentives corrupts design judgment is an accurate description of current training, not an argument against integration. An architect who carries financial risk has a direct incentive to understand what makes a building perform over time, retain value, generate community attachment, and serve its users. A developer who carries design accountability has an incentive to understand what makes buildings legible, durable, and socially embedded. These interests are not inherently opposed. They are opposed because the professions are structured to optimise for their own part of the process, with no accountability for the whole. Integration does not corrupt the design agenda. It replaces the current arrangement, where the design agenda is already corrupted, with one where both parties are accountable to whether the building actually works.

The obvious objection is that time spent raising capital and financing construction is time not spent doing architecture. Specialisation produces excellence, and diluting focus across two demanding disciplines potentially produces mediocrity in both. There is a stronger version of this objection too: the architect who controls financing is no longer independent. They now have a direct financial stake in the outcome, which creates exactly the conflict of interest that currently corrupts developer-driven architecture. The profit motive does not disappear when the architect holds the capital. It just moves closer to the design desk.

Move it to the first paragraph, where it serves as the conclusion of the leverage argument rather than an orphaned line in the second. Like this:

It is a reasonable position, and I held something like it for a long time. But it assumes that the current arrangement actually allows architects to do good architecture sustainably. Does it really? The reason architects spend their careers negotiating over finish specifications, absorbing value-engineering cuts, compromising programme decisions, and remain among the most lowly paid of regulated professions, is precisely because they control none of the capital and therefore almost none of the leverage. Getting better at design within that constraint makes you a more skilled servant of someone else’s agenda. It does not make you more capable of determining what gets built, for whom, or why. The conflict of interest argument, meanwhile, proves too much. Architects already operate under conflicts of interest: they serve clients whose interests routinely diverge from those of building users, future occupants, and the public. The current arrangement does not protect design independence. It just ensures that the compromising happens on someone else’s terms rather than your own.

What the current system actually rewards is not design excellence. It rewards the willingness to treat architecture as a business development tool: to compromise on design in service of capital, to move fast, to absorb the client’s agenda without friction, and to produce buildings that satisfy the brief and churn profits, rather than satisfy the people who will live and work in them. Those are the practices that win the commissions, keep the lights on, and grow the firm. The question worth sitting with is a simple one: if the built environment is shaped primarily by those willing to subordinate design to financial return, is that the built environment we want to live in? The choice is not between doing architecture and doing finance. It is between accepting the current terms of practice or taking a bet on building the conditions under which better architecture becomes possible in the first place.