IN Brief:
- Artificial intelligence is now used by 74% of surveyed architecture practices, up from 59% in 2025.
- Three-quarters of users report productivity gains and 57% identify a positive return on investment.
- Only 17% report better design quality, while concerns persist over jobs, training, intellectual property, and accountability.
Artificial intelligence is now used by 74% of architecture practices surveyed by the Royal Institute of British Architects, marking a further acceleration in the profession’s adoption of generative and analytical tools.
The proportion has risen from 59% in 2025, with applications extending from early-stage visualisation and option development to project management, administration, research, bid preparation, and routine practice operations.
Three-quarters of respondents using AI reported improved productivity, while 57% identified a positive return on investment. Adoption is consequently moving beyond informal experimentation as more practices measure whether software subscriptions, training, governance, and workflow changes produce a commercial return.
Design outcomes remain less conclusive, with only 17% of respondents saying AI had improved design quality. The gap between faster production and better buildings remains substantial, because generating images, text, schedules, or options more quickly does not necessarily strengthen technical resolution, coordination, or spatial judgement.
Employment and professional development also feature prominently in the findings. Fifty-nine per cent of respondents expect AI to reduce staffing requirements, while 61% believe it will make it harder for people entering the profession to acquire the skills needed at later career stages.
Junior architects and technicians have traditionally learned through drawing production, research, modelling, specification work, coordination, and repeated review by senior colleagues. Automating too much of that foundation work may improve short-term utilisation while narrowing the range of tasks through which professional judgement develops.
Practices therefore face a workflow-design challenge alongside the choice of software. Tasks suitable for automation must be distinguished from work that should remain visible to trainees, checked by competent professionals, or retained because it builds understanding of structure, materials, regulation, procurement, and construction sequencing.
AI is also entering adjacent construction workflows, including mechanical and electrical estimating systems that use automated take-off functions. As the technology moves into cost and procurement activities, inaccurate outputs can affect commercial decisions as readily as design development.
Connected systems can allow an early design assumption to pass rapidly into quantities, cost plans, embodied-carbon estimates, programmes, and tender information. When the source data is reliable, the efficiency gain can be considerable; when it is wrong, automation may distribute the same error across several disciplines before its origin is recognised.
Professional responsibility remains with the architect and appointed organisations rather than the software. AI-generated material must still comply with contractual obligations, building regulations, planning requirements, copyright law, information-management procedures, and the practice’s duty to exercise reasonable skill and care.
Defined review points are consequently essential. Practices need to establish which systems may be used, what information can be entered, how outputs are identified, who checks them, and whether prompts, source data, model versions, and approval records must be retained.
An attractive image or fluent technical note can appear authoritative even where important assumptions are missing. Experienced review becomes more important as the software becomes more convincing, because poor-quality outputs may no longer look obviously incomplete.
Intellectual-property risk remains another unresolved area. Project information may include confidential client material, designs owned by several parties, personal data, commercially sensitive costs, or details of secure buildings.
Uploading such information to an external system without understanding storage, retention, and model-training arrangements can breach appointments or data policies. Practices also need to know whether generated material draws on copyrighted sources and whether the resulting output can be used commercially without challenge.
Procurement use demands similar caution. AI may help compare submissions, identify missing information, or organise evaluation material, but scoring decisions must remain transparent, explainable, and consistent with the published process.
A system unable to show how it reached a conclusion can expose a client to challenge even where its recommendation appears reasonable. Human review must cover both the outcome and the basis on which it was produced.
Environmental performance also warrants scrutiny. AI can support energy modelling, option testing, material comparison, and design optimisation, although large models carry their own computational and energy demand. Digital sophistication cannot be treated automatically as an environmental benefit.
The strongest gains are likely to come from tightly defined applications connected to reliable project data. Searching standards, checking model information, classifying documents, drafting repetitive schedules, identifying coordination issues, and generating controlled options all have clearer boundaries than asking a general-purpose system to produce an apparently complete design response.
Only 21% of surveyed architects were more optimistic about AI than a year earlier, despite the rise in adoption. The profession appears to be using the technology pragmatically, as measurable productivity gains develop alongside a clearer view of operational, legal, and employment risks.
Governance, competence, and training will shape the next phase. Practices that allow unmanaged individual use may gain speed but create inconsistent records and hidden liability, whereas controlled workflows offer a better prospect of retaining productivity without weakening professional oversight.
Adoption is likely to rise further, but the more useful measure will be whether faster production leads to better coordinated, compliant, and buildable information. Architecture practices will still be judged through the quality of their professional decisions, regardless of how much of the underlying work has been automated.
