An architecture firm in Portland recently told a prospective client that their design process had been "transformed by artificial intelligence." They showed renderings generated in minutes instead of days. They demonstrated concept iterations that would have taken a junior designer a week. Impressed, the client signed.
That fee structure was identical to every contract the firm had written since 2019: schematic design at 15%, design development at 20%, construction documents commanding the lion's share at 40%, bidding and negotiation at 5%, and construction administration at 20%. These percentages are not unique to that Portland firm. They are the standard phase breakdown codified in AIA Document B101, the contract template that governs the overwhelming majority of residential and commercial architectural engagements in the United States. They have been essentially unchanged for decades.
That number lands differently when you set it beside another one.
Sixty-four percent experimenting with the tools, ninety-seven percent unchanged in the process that governs how those tools get used, billed, and delivered to clients.
The waterfall that will not bend
The AIA's five-phase structure originated in an era when architectural deliverables were physical objects. Schematic design produced hand-drawn sketches on trace paper. Design development refined those sketches into dimensioned drawings. Construction documents translated them into blueprints that could be photocopied and handed to a general contractor. Each phase existed because each phase produced something tangible that the client could hold, review, and approve before paying the next installment.
AutoCAD arrived in 1982 and did not change the phases. Revit arrived in 2000 and did not change them either. BIM became industry standard by 2015, and still the phases held. Now AI generates concept renderings in seconds, iterates floor plans against zoning constraints in minutes, and produces photorealistic visualizations that once required days of manual compositing, and yet the phases remain exactly as they were when every deliverable was a rolled-up sheet of trace paper.
Where AI actually lives in this process
Chaos and Architizer's survey is specific about where architects find AI useful: forty-three percent identify concept and pre-design as the area of greatest impact, image generation and enhancement dominate these early-stage workflows, and eighty-five percent report time savings concentrated in the stages where speed and iteration matter more than precision.
This is telling.
AI has colonized the schematic design phase, the part of the process that represents 15% of the total fee. It has done almost nothing to the construction documents phase, which represents 40%. Construction documents, representing 40% of the fee, remain untouched by the technology that supposedly transformed the profession.
When a firm tells you AI makes them more efficient, ask a simple question: did your fee go down? In nearly every case, the answer is no, because that Portland firm's AI-generated concepts replaced work that took a junior designer a week, but that designer still works there, the reclaimed week gets allocated to other tasks, and your invoice remains the same, broken into the same phases, paid on the same schedule, covering the same scope of services defined by the same AIA contract.
The contract is the constraint
This is not a failure of technology. It is a failure of imagination, reinforced by contractual inertia.
AIA Document B101 defines basic services by phase. Insurance policies tie professional liability to phase-based deliverables. Payment milestones correspond to phase completions. An architect who wanted to restructure their workflow around AI capabilities would need to rewrite their standard contract, renegotiate their E&O coverage, and convince clients to accept a payment structure that doesn't map to the familiar five-phase model their lender, their contractor, and their building department all expect.
Nobody is doing this, as the survey data makes clear: only 20% of architects who experiment with AI have "fully embraced" it in their workflow. And even that 20% operates within the same phase structure as the other 80%. Full embrace, in practice, means running Midjourney during schematic design instead of sketching by hand. It does not mean questioning why schematic design exists as a separately billable phase in the first place.
The 25-month contract decline
There is a darker reading of these numbers.
AIA's March 2026 report reveals that the value of newly signed design contracts has decreased for 25 consecutive months, with architecture firm billings essentially flat at an ABI score of 49.8 and backlogs rising to 6.6 months, the highest since December 2023, which means firms are working through old commitments rather than signing new ones at comparable rates.
During this same 25-month decline, AI experimentation in architecture roughly doubled from around 30% to 64%, time savings claims rose to 85%, and yet firms got faster without getting busier, without getting cheaper for clients, and without translating any of those efficiency gains into market advantage. Efficiency gains were absorbed internally, converted into margin or redistributed across existing staff, while the market continued to contract.
For homeowners commissioning custom residential work, this means the AI productivity your architect advertises has not translated into lower fees, faster timelines, or demonstrably better outcomes. It has translated into better-looking Instagram posts during the concept phase.
The small firm squeeze
Sixty percent of the Chaos/Architizer survey respondents work in firms with fewer than 20 people. These are the practices most likely to design your home. They are also the ones most constrained by traditional phase structures, because small firms rely on AIA contracts as protective scaffolding. A three-person studio cannot afford to invent a bespoke service agreement for every project. Standard phases give them a framework for scoping work, managing client expectations, and defending themselves in disputes.
AI tools promise these firms disproportionate productivity gains. A sole practitioner generating concepts with AI can compete visually with a 50-person firm during the pitch. Once the contract is signed, both firms enter the same waterfall. A sole practitioner still needs to produce design development drawings, construction documents, and field administration under the same phase-based fee structure. AI helped them win the job. It did not help them do the job differently.
Satisfaction data from the survey reflects this disconnect: sixty-nine percent of architects report being only "somewhat" satisfied with AI output, and forty-eight percent cite inconsistent or poor quality as their biggest obstacle, complaints that signal a technology usable in one corner of a process that has five.
The strongest case for doing nothing
There is a defensible argument that 97% traditional phase adoption is exactly right.
Phases exist because buildings are expensive mistakes. A $500,000 custom home with a structural error in the construction documents can generate liability that destroys a small firm. Phase-gate structures force incremental review, client sign-off, and documented decision points that protect both parties. Removing or compressing phases in the name of AI efficiency could eliminate the checkpoints that catch errors before they become change orders, claims, or worse.
Insurance carriers price E&O premiums based on established workflows. An architect who restructures their process around AI-generated deliverables may find themselves in uncharted territory with their insurer, particularly after Verisk's 2026 AI exclusion endorsement forms gave carriers new tools to limit coverage for AI-assisted work.
And clients understand phases. They know what they are paying for at each milestone. A homeowner reviewing schematic design drawings feels oriented in the process. Replacing that with "we ran 400 AI iterations and here are the three best ones" may be more efficient but is less legible to the person writing the checks.
These arguments are real, and they are also the arguments that kept the profession on hand drafting for a decade after AutoCAD was available, and on 2D workflows for a decade after BIM was viable. Reasonable caution and structural inertia are difficult to tell apart from the outside.
What this means if you are building a home
If your architect mentions AI during the sales process, treat it as a question, not a credential. Specifically:
- Ask which phases use AI. If the answer is only schematic design or concept visualization, you are paying a traditional fee for a traditional process with a faster first step. That is fine. It is not transformative.
- Ask whether AI changed your timeline. If the overall project schedule from first meeting to construction documents is the same as it would have been three years ago, AI did not save you time but saved the architect time, and they kept the difference.
- Ask about the fee structure. If the phase percentages are 15/20/40/5/20 or close to it, you are paying the same fee breakdown that predates AI, BIM, and AutoCAD, which does not mean the fee is wrong but means the "efficiency" claim deserves scrutiny.
- Ask what happens to AI-generated concepts in design development. Many firms generate dozens of AI concepts during schematic design, then manually redraw the selected option in Revit for design development. AI-generated concepts get thrown away and rebuilt in a traditional tool. If that is your architect's workflow, AI is a presentation trick, not a design method.
Limitations of this analysis
The AIA's 97% figure comes from a survey question about whether firms "still work within traditional design phases." It does not break down whether firms have modified those phases internally while preserving the external framework. Some firms may run AI-intensive processes within a schematic design phase that looks traditional on the contract but operates very differently in practice. That headline number captures contractual structure, not necessarily internal workflow.
Chaos/Architizer's survey skews toward visualization-heavy practices, since Chaos sells rendering software. Firms whose AI use centers on structural analysis, energy modeling, or code compliance may be underrepresented. Self-reported time savings of 85% lack third-party validation. And no dataset currently tracks whether AI adoption has changed fee percentages at individual firms over time, which would be the most direct measure of whether efficiency gains reach the client.
International variation matters here: European firms operating under different contractual frameworks may have more flexibility to restructure phases, and this analysis focuses specifically on the U.S. market and AIA contract norms.
The building doesn't care
A house does not know how it was designed. No framing contractor cares whether the plans originated from an AI concept generator or a hand sketch on a cocktail napkin. Building inspectors review construction documents for code compliance regardless of how many AI iterations preceded them.
What AI promised architecture was never just speed but structural change in how buildings get designed, reviewed, and delivered. Four years in, the speed arrived while the structural change did not, and the phases and the fees persist unchanged. The process your architect follows in 2026 would be immediately recognizable to an architect from 1996, with better renderings.
That is not a technology problem. It is a profession that adopted a powerful new tool and used it to do the same thing, slightly faster, for the same price.
Your invoice will reflect the difference.