Architects thinking about AI generally fall into one of two camps:

• Camp 1: How AI can be leveraged to make current workflows faster, more accurate, and less expensive.
• Camp 2: How AI can be used to accomplish objectives by circumventing or obviating existing workflows.

I get lots of questions about the former, but actively try to avoid answering them. The problem is that Camp 1 thinking presumes current practices are worth amplifying and that workflows conceived under one technological era will remain optimal under a subsequent one. That’s not always the case, and can often prove to be counterproductive.

Consider the QWERTY keyboard. In early typewriters, typing at excessive speed would jam the mechanical keys. Manufacturers deliberately designed a difficult keyboard layout to slow typists down, placing commonly used letters far apart. When computers eliminated the jamming problem, we kept the inefficient layout simply because that’s what everyone had grown used to. Consequently, rates of Carpal Tunnel Syndrome continue to climb because we failed to question whether a design created for obsolete constraints should continue into a new technological era.

If only someone at the dawn of the computer age had had the good sense to ask whether we should continue using the QWERTY keyboard layout. Yes, a transition would have been tough as everyone would learn how to type completely differently. But you know what else is tough? Carpal Tunnel Syndrome.

Architecture stands at a similar inflection point with AI. Rather than shellacking AI onto our existing methods, we have an opportunity—a mandate, actually—to fundamentally rethink how architectural services are delivered in an AI-native world.

Current AI “Choices” Probably Aren’t Choices at All

New technological regimes create a window for asking fundamental questions about how we do work and why we do it the way that we do. Sometimes these questions barely get asked at all, as the QWERTY keyboard history demonstrates. Other times these questions get asked and then answered in revolutionary ways, as seen with the “digital first” revolution in journalism. 

Any older journalist would today claim that the transition in journalism from print to digital wasn’t a choice at all, but a mandate thrust upon the profession by the rise of the internet age. That’s obvious from the viewpoint of the present. But at the time it was considered a choice, one hotly debated and faced by every newsroom.

I humbly submit that every architecture firm now faces a similar choice about whether to be “AI first.” As with the “digital first” movement in journalism, it’s the kind of “choice” that will be revealed by history to have not been a choice at all.

Remembering “Digital First”

“Digital first” emerged from the journalism world as publications grappled with the internet revolution. Traditional print media operated on rigid schedules: reporters finished stories by 7:00 p.m. because editors needed to finish by 8:00, layout by 10:00, and printing by 11:00, all so distribution could begin the next morning at 4:00 a.m. This structured rhythm shaped everything about newspaper operations, from staffing to workplace design to bureau locations.

The constraints of this model determined when readers received their news: if a major story broke at 9:00 p.m., it waited until the next issue of the newspaper. In delaying, publications didn’t need to worry about getting scooped, because all of them operated under the same constraints.

Digital publishing eliminated this physical infrastructure and its associated constraints. A digital-first publication could:

• Publish at any time instead of adhering to print schedules.
• Release breaking news immediately, without space limitations.
• Incorporate multimedia elements impossible to replicate in print.
• Continuously update stories continuously as developments occur.

Media outlets faced a fundamental choice: remain print-first, transferring print versions to a digital format after publication, or become digital-first, optimizing for the new medium. The decision would eventually impact staffing structure, workflow design, revenue models, and competitive positioning.

To make that decision, a paper had to internally recognize that the primary objective—delivering the news—remained unchanged, but the optimal methods for achieving it had been completely transformed by new technologies.

Consider: Preindustrial traders crossed mountain ranges via established paths suitable for horses. Twentieth-century engineers didn’t follow these routes, instead tunneling through the mountains where the geology and geography were most favorable. Today, airplanes ignore the terrain entirely and take the route that is optimal for fuel economy. 

Camp 1 thinking may seem prudent today, especially if you’re running behind on a project for a difficult client. But it’s the logic of a civil engineer who designs his train track to follow the old horse trail. Camp 2 thinking asks: Now that we have this new technology, what does it say to us about new possible paths?

How Architecture “Delivers the News”

That’s all very lofty and generic, but what would an AI-first architecture firm actually look like? We can begin by examining component workflows in architecture and asking ourselves how we would approach them in an AI-first world.

Example 1

• Traditional Project Phasing: Having phases of SD, DD, and CD where drawing packages are produced and reviewed, either internally, externally, or both.
• AI-First Project Phasing: Having a system of continuous review, performed by a central firm intelligence, that flags any design changes that might eventually exceed budgetary, energy, or programmatic constraints. A central firm intelligence could be fed by an ongoing data stream of notes, meeting minutes, evolving price data, etc., which would ensure that it is up to date, in real time, on all of the project’s hard and soft requirements. 

Example 2

• Traditional Client Communication: Having regular check-ins with the client where design teams seek approval on certain design decisions or input on certain project requirements.
• AI-First Client Communication: Embodying the client as an AI avatar that can be asked any question, at any time, at the architect’s discretion.

In both examples, the architecture firm is still “delivering the news.” Both project phasing and client communication basically have the same ultimate objective: to keep the designer’s efforts aligned with the client’s wishes (even when those wishes are dynamic and mercurial). Both approaches accomplish this, but in the second case, rather than using AI to simply amplify or accelerate existing practices (e.g., I will use AI to respond to client emails), the architects are using the technology to achieve the objective—compliance with clients’ intent—by a wholly different route. 

The same logic can be scaled upward to the size of a firm.

An architecture firm’s primary objective is to provide design services to clients who are interested in developing those designs into built works. The manner in which an architecture firm delivers on this principal objective should provide a reasonable assurance that under the given constraints, the final building design represents the optimal solution for the client’s goals, done in accordance with a widely held standard of care.

If we were to pursue that optimal solution, with no attachments to any of the practices, procedures, and methods on which we were trained, and no respect for any of the architectural customs of the past, using all of the AI-enabled technology that we have today (or will have in the near future), how would we do it?

Pursuing any objective toward an optimal solution usually begins with breaking the objective into smaller, more manageable objectives that are aligned to that central goal. For an architect, it’s not so simple. Architects juggle multiple, overlapping, and sometimes contradictory objectives. Optimizing for any one of them could easily negatively impact another. There are mathematical techniques (e.g., evolutionary algorithms, Pareto optimizations) that can make short work of a multi-objective optimization problem like these. But for them to work, one would have to be able to define the terms of optimization. How does one “optimize” client satisfaction? By giving them a survey? An fMRI? What if the client is just a curmudgeon?

This is probably why architecture firms have resisted the kind of “McKinsey-ification” that has happened to most of corporate America. A productive, successful, and well-run firm might try to optimize for:

• Number of clients
• Number of happy clients
• Client satisfaction
• Number of projects
• Quality of design
• Number of design awards
• Number of employees
• Employee satisfaction
• Profit

But there’s no straightforward way to optimize the firm, for two reasons:

• Each of these objectives has a relationship with the others. It could be a complementary relationship (client satisfaction may boost profitability) or an inverse relationship (optimizing for number of projects may push down design quality).
• Most of these objectives resist straightforward quantification. For example, you can count the number of design awards received, but not all design awards are equivalent.

These may be the sort of mysteries that remain solely within human scope for a while. Until we can tell a machine what a “happy” client actually looks like, the machine will stand by, and architects can make their own judgments about what an “optimal” firm looks like. That is probably for the best. Different architects, taking different positions on what “good” architecture is, and how to do it, is one of the things that gives our built environment diversity, intrigue, and delight.

Adapting Now for an Uncertain Then

The problem is that the AI revolution is happening now. If we’re committed to making changes in our practice that optimize for success in the age of AI, then we have to act. But acting on our current workflows leads us down the old horse trail. We will, by default, land in Camp 1, trying to apply AI atop workflows conceived in an ink-and-vellum era.

In sum, we must position ourselves now to take advantage of this new technology, while optimizing on multiple, competing objectives, and also avoiding the allure of all the methods that made us successful in the first place. How in the hell does anyone do that? What do we even optimize for at that point?

Simple: To build an AI-first firm, we would optimize for the technology itself. We organize the firm in such a way that it maximally positions AI to help us in whichever goals we eventually pursue, and however we eventually define criteria for success.

“Optimizing” for AI doesn’t mean having the most AI, or the latest LLM model. It means organizing the structure of the firm around key, evergreen principles that allow AI systems to do their best work and remain flexible enough to accommodate the fact that AI is advancing rapidly.

The Five Optimizations

Optimizing for AI won’t guarantee success in the AI era. But not optimizing for AI will almost certainly guarantee failure. Many of the digital-first publications of the 1990s went out of business, and many of the old print publications successfully navigated the transition. But no major publication today is print-first, or print-only.

To optimize for AI requires five distinct sub-optimizations, which I have taken to calling, quite cleverly, “The Five Optimizations”:

1. Optimize for Fast Institutional Learning. Pre-AI, the “wisdom” of a firm was the accumulated wisdom of its people. Wisdom comes from experience. In architecture, experience comes from projects—and projects take a long damn time. In a world where technology evolves at exponential rates, the organization itself must learn and adapt faster than any individual within it. This requires fundamentally different knowledge-sharing architectures than those in place today, which depend too much on principals and senior architects serving as the stewards of firm knowledge. AI-first firms would build new systems of constant, omnidirectional learning, utilizing AI as a hub that can surface appropriate information, to the appropriate person, at the appropriate time.
2. Optimize for Microscale Innovation. AI will fundamentally democratize innovation and enable individual architects to innovate rapidly at small scales. The traditional innovation model of relying on a few key tech-savvy individuals in one corner of the firm gives way to an expansive ecosystem of constant micro-improvements. Within AI-first firms, architects will leverage AI to create such innovations. And at the firm level, the firm will leverage AI to systematically capture them and intelligently disseminate them across all projects. Simultaneously, AI-first firms will develop new methods of compensation that are less about rewarding labor and more about rewarding the innovations that eliminate labor.
3. Optimize for Demonstrable Social Value. As AI increasingly commoditizes technical expertise, the value proposition of architecture firms must expand beyond the direct client relationship. Firms need systematic approaches to demonstrate broader societal impact, making their value visible to clients, communities, and collaborators alike.
4. Optimize for Data Consciousness. The coming tsunami of architectural data—from climate models to user behavior patterns to material performance metrics—will overwhelm traditional data-handling capabilities. Successful firms will avoid trying to monetize their historical data, which will largely be value-less, but will instead cultivate a data consciousness that allows them to sculpt future architectural practices in ways that turn design services and the buildings they produce into data-harvesting machines. AI-first firms will work to capture the value of those data streams in negotiation with their clients.
5. Optimize for Blurry Planning. The accelerating pace of technological change renders traditional planning methods largely obsolete. AI-first firms operate with “blurry planning”—maintaining a clear strategic direction while employing flexible tactical approaches that can rapidly adjust to emerging technologies, shifting market conditions, and evolving client needs. The state-of-the-art AI at the end of your project may be 1,000 times more powerful than the state-of-the-art AI at the beginning of your project. Successful AI-first firms will adapt, in their operations and culture, to overdriving their headlights. Adopting and optimizing for blurry planning will prove more challenging in architectural services than in most industries because of the industry’s long time scales. As such, even AI-first architecture firms will not be able to do the hard pivots that are popular in the startup world. Instead, they will cultivate hybrid planning strategies that allow for steady progress as well as conditional, tactical pivots.

Such a list will provoke some obvious questions, such as, Why not optimize for profit, design awards, or something similar? These would seem to be the things a firm would want the most of. However, we can’t optimize for a thing when we don’t yet understand how it’s made or achieved. If we wanted to optimize for design awards, does that mean hiring all the best human designers, so that they can stay a step ahead of AI? Or does it mean cultivating in-house AI design assistants for one’s existing staff? Does it mean subscribing to a SAAS-type AI design model that Autodesk will certainly produce at some point? Or does it mean building your own? All of these questions face the same disturbing reality: We don’t know how fast this technology will advance, nor what it will advance into. And if we wait until we are certain about those things, it will likely be too late to make the sort of internal changes needed.

Any existing firm principal has some strategy to optimize for profit (or design awards, or whatever) in a non-AI world. But no one has any idea whether those same strategies will be optimal in an AI future. Therefore, we optimize for the things that are going to allow us to maximally adapt to a future where the only thing we know for sure is that AI will be as pervasive as it is powerful. That will put us in a position to take the best advantage of the technology in whichever direction it eventually develops.

The most important first step is in deciding to be AI-first. As it was with journalism, this initial decision will provoke the fundamental questions about the financial, strategic, organizational, and design strategies needed in the AI era. Every firm will answer them differently, and that will be a good thing. Collectively, we’ll then be able to answer the larger questions that AI poses to architecture.

Featured image created by the author using AI. Visit the author’s Substack and subscribe for free.