The numbers look impressive until you read the fine print
Architecture has always absorbed new technologies on its own terms. CAD took decades to displace the drawing board. BIM adoption stretched across an entire generation of practitioners. AI is moving faster, but the pattern of uneven uptake is remarkably familiar.
Phil Bernstein and Vincent Guerrero, writing for RIBA Journal in March 2026, identified four areas where AI is developing rapidly within architectural practice: workflow automation in larger firms, building engines for structural engineering, LLM-based code generation for computational design, and broader adoption filtering into smaller studios.
Their analysis, rooted in conversations across AEC practices, paints a picture of an industry that has moved past curiosity but has yet to arrive at strategic integration.
The RIBA AI Report 2025 quantified the shift. Practices using AI rose from 41% in 2024 to nearly 60% in 2025, with large practices exceeding 80% adoption rates. Smaller studios sat at 48%. The most common applications centred on early design visualisations and specification writing, tasks where efficiency gains are immediate and the risk of error is manageable.
Excitement is outpacing governance
At RIBA's AI in Practice Summit in February 2026, the mood was instructive. A show of hands revealed that most delegates used AI daily, and some used it hourly. The question had shifted from whether architects should use AI to how they should govern it.
Fewer than one in five practices have invested in research and development around AI, and only 15% have formal AI policies in place. That gap between experimentation and structured implementation matters. As Beale & Co observed in its analysis of the RIBA report, the profession's embrace of AI is being driven by pressing challenges around sustainability and performance, but the governance structures are trailing behind the tools.
Hamza Shaikh, creative AI lead at Gensler, warned at the Summit against focusing too heavily on individual tools at the expense of a coherent approach. He advocated for putting "our human craft at the forefront" of AI adoption, and anticipated new practice roles emerging, particularly around data organisation and interoperability. Firms should already be recruiting staff with these skills, he argued. The spectre of "software exhaustion" loomed large in several presentations. When every week brings a new plugin or platform, the risk of fragmented workflows is real.
The design companion framing is winning, for now
More than nine in ten architects surveyed by RIBA rejected the notion that AI could substitute for professional decision-making. The consensus view is that AI should function as a "design companion," augmenting creative judgement rather than replacing it. Autodesk's Forma Building Design and deeper Revit cloud integration are enabling teams to explore design options earlier in the project lifecycle, with greater continuity between concept and delivery.
Saina Abdollahzadeh, lead architect at Studio Tim Fu, reinforced this position at the Summit: "The slops are mostly when you let AI do the design and take you over."
The useful applications, she contended, sit firmly in the territory of assistance and agency, with the architect maintaining control.
That framing is pragmatic and defensible. Whether it holds as AI capabilities accelerate is another question entirely. Stanford Anderson's 1966 lecture at MIT, which Bernstein and Guerrero reference in their RIBA Journal piece, described architectural design as a process of "worrying" about a problem rather than solving it algorithmically. Six decades later, the worry has simply found new tools.
Smaller firms face a different calculus
The adoption gap between large and small practices is more than a resource issue. Large firms can absorb the cost of experimentation and dedicate personnel to integration. A five-person studio evaluating AI tools is making a bet with a much higher proportion of its overhead.
The RIBA data shows smaller studios at 48% adoption, which is notable progress, but the nature of their usage tends to be narrower. Administrative automation and visualisation are entry points. More advanced applications, such as performance simulation or environmental modelling, remain concentrated in firms with dedicated computational design teams. The risk is that smaller practices adopt AI for efficiency gains while missing the more transformative opportunities in design intelligence and client engagement.
The question nobody answered at the Summit
An audience member at the RIBA Summit asked whether AI was helping to build better buildings. The panel could not offer a definitive answer. Tomas Millar, architect and founder of Millar + Howard Workshop, captured the prevailing sentiment: "No one knows where this is going."
That honesty is refreshing. The architectural profession is navigating a period of genuine uncertainty, and the temptation to project confidence about AI's trajectory should be resisted. What we can observe is that adoption is accelerating, governance is lagging, and the profession's long-standing emphasis on creative judgement is being tested by tools that can generate plausible design outputs in seconds.
The next twelve months will reveal whether the "design companion" framing can survive contact with models that are increasingly capable of independent design reasoning. For project delivery professionals watching from adjacent disciplines, the architectural profession's struggle with AI integration offers a preview of tensions that will soon ripple through every stage of the construction lifecycle.
Takeaway
Practices without formal AI policies risk adopting tools without the governance structures needed to manage liability, data security, and intellectual property. The 15% with policies in place have a structural advantage that will compound over the next year.
The emergence of new roles, such as AI integration specialists, signals that firms treating AI as a plug-in rather than a strategic capability will fall behind those building dedicated capacity.
Smaller studios should evaluate AI adoption through the lens of client-facing differentiation, not internal efficiency alone. The firms that use AI to offer faster concept iteration or more rigorous performance analysis will win work from those who use it to cut overheads.
The inability to answer whether AI is producing better buildings is itself a finding. Until the profession develops metrics to evaluate AI's impact on design quality, adoption will remain driven by speed and cost rather than outcomes.
If this piece sharpened your thinking on AI in AEC, subscribe to Project Flux for weekly analysis that cuts through the noise. We cover what matters for project delivery professionals, every week.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
/
