Subscribe
Share
Share
Embed
Reimagining AEC is Dr Sepehr Abrishami's independent podcast on how the built environment handles information. Each episode takes one subject from digital construction, from the ISO 19650 series and the common data environment to the golden thread, soft landings and the arrival of artificial intelligence in design and surveying practice, and tells it as a single argument built on the published standards, guidance and evidence. Sepehr is a researcher and former architect who leads the Centre for AI in Construction and teaches building information modelling at master's level. The episodes are written for postgraduate students and practitioners who want the reasoning behind the rules, not a summary of them.
The question in this episode is how far artificial intelligence, or AI, has moved from trial into routine practice in construction and commercial property. I am Dr Sepehr Abrishami, and because the centre I lead is the Centre for AI in Construction, I cannot treat that question as settled. So on Reimagining AEC I take it to the evidence, starting with a warning in the report this episode is built on. AI, it says, reaches recently qualified professionals when they have the least experience to judge what it produces.
That gap in experience now meets a written duty. Under a professional standard from the Royal Institution of Chartered Surveyors, or RICS, a regulated firm that runs a client's contract documents through an AI tool cannot simply use what comes back. Someone qualified, and named, has to put in writing whether the tool's answer can be trusted for the service the client is paying for. A rule like that assumes AI has become part of everyday professional work. In August 2026, RICS published survey evidence on that very point, and its foreword makes a strong claim. Maureen Ehrenberg, President Elect of RICS, writes that a year earlier the report described an industry approaching an inflection point. In her words, this year's data shows that point has been crossed.
Crossed could mean that AI is now routine in professional work. It could also mean only that most firms have started to try it. Which reading the evidence supports is what this episode settles. The aim is that you can read the survey evidence with its limits in view, explain why so many pilots stop short of routine practice, and apply the standard's main requirements to one pilot. That pilot belongs to the running example, a cost consultancy regulated by RICS that is testing an AI tool to review contract documents. The consultancy is my invention. The figures and requirements it meets here are real. I begin with how the survey was built, follow adoption and the barriers behind it, and apply the standard to that pilot. I end with how far firms say they meet the standard and what the report asks of industry and RICS next.
Who answered comes first. The source is the AI in commercial property and construction report 2026, by Tom Storey, Anil Sawhney and Katherine Pitman, published by RICS. Its questions ran inside two regular market surveys that RICS conducts, the Global Construction Monitor and the Global Commercial Property Monitor. Together they produced 3,148 responses in the first quarter of 2026, 1,883 from construction and 1,265 from commercial property. Construction can be compared with the 2025 edition of the report. For property, this is a first baseline.
Four features limit how widely the results apply, and they reduce to two concerns. The first is who is counted and how. The respondents answer RICS monitor surveys, so they describe a professional community close to RICS, not every firm in either sector. The report says UK results were weighted at nought point nought four and the rest at nought point nine six. It also says the global results average five regions equally. It does not explain how the two rules combine, and it reports no response rate and no test of statistical significance. The second concern is what an answer is. Every answer is self-reported, an account of an organisation's AI use that no one has observed. And the two sectors were asked similar but not identical questions. Every percentage that follows inherits both concerns.
The adoption evidence comes in four levels. The lowest is some use of AI anywhere in the organisation. Around two thirds of construction respondents now report it, up from just over half in 2025, and the share reporting no AI use at all fell by over a third. In commercial property, more than three quarters report some use. The second level is the early-stage pilot, a trial of a tool on limited work. It is the most common position in both sectors, at an identical 39 per cent. The third is regular use in specific processes. It rose from 12 to 19 per cent in construction and stands at 29 per cent in property. The highest, widespread or full use, is rare, at 4 per cent in construction and 6 per cent in property.
Two cautions follow. The three upper levels add to 62 per cent in construction and 74 per cent in property, so read the phrases around two thirds and more than three quarters as generous roundings. And the UK, which started as the most cautious region, remains so, although it recorded the largest fall in no AI use of any region. Our cost consultancy sits at the second level, one of the 39 per cent running a pilot.
The report argues that the gap in adoption has moved. In 2025 it lay between planning to invest and starting. In 2026 it lies between running a pilot and embedding it in routine work. Both sectors pilot at the same rate, so the difference between them is conversion. The authors offer two contrasts to explain why property converts more of its pilots. One concerns data. Commercial property data sits largely in portfolios held by one firm, so a tool that works once can keep working across many assets. Construction data resets at each project boundary and comes from many firms working to different standards, so the effort of getting a tool running must be repeated. The other is income, which recurs across asset cycles in property but arrives job by job in construction, so each investment there must prove itself on a single project. These are the authors' explanations, not results the survey tested. For our consultancy, a successful pilot on one project proves little until the firm has planned how the tool will meet the next project's documents.
Why so many pilots remain pilots is the next thing to settle, beginning with what respondents expect AI to do. Because the two surveys offered different options, compare the shape of the expectations rather than their rankings. Construction respondents chose design optioneering, in which AI generates design options for a professional to judge, most often for impact over the next five years, although its share fell from 2025. Regulatory compliance was among the next most selected areas, and UK respondents chose it over twice as often as those elsewhere. The report links that to the gateway regime under the Building Safety Act 2022, to new energy standards for homes and other buildings, and to carbon reporting, which have all added reporting work. The gateways are the subject of the episode on the golden thread for higher-risk buildings. Property respondents favoured decision support for valuation, market analysis and portfolio choices. Compliance sits near the top of both lists.
At the bottom of the construction list, safety and sustainability rank lowest, both for how significant respondents think AI could be and for its expected impact over five years, as they did in 2025. Yet these are the areas where professional and regulatory obligations are rising fastest. The report suggests two reasons. Safety-critical AI needs integrated data, real-time supervision and clear accountability, which are not yet consistently in place. And the accountability frameworks around these outcomes have not yet adapted to AI.
The construction barriers changed little in a year. Lack of skilled personnel still heads the list, and the top three barriers are the same as in 2025. Two movements stand out. Concern about privacy and security rose from 22 to 30 per cent. The report tentatively links this to rising use, as firms using AI more meet data handling questions. The European Union's Artificial Intelligence Act, in force since August 2024 with obligations applying in phases, has made those questions more visible. Our consultancy meets this barrier the moment a client's documents go into its tool. High implementation costs moved the other way, from 29 to 25 per cent, as trialling AI became cheaper and AI features arrived inside software that firms already paid for. Commercial property cites data quality and availability first, integration with existing systems second, and skills only third. The report reads these as the barriers of a sector already deploying at scale, although only 6 per cent there report widespread or full use.
Spending plans are best paired with present capability. In construction, 37 per cent plan a moderate or significant increase in AI investment over the next 12 months, up from 30 per cent a year earlier, yet only 4 per cent report widespread or full use. In commercial property, 47 per cent plan an increase, against 6 per cent in widespread or full use. These are different measures, so the comparison shows a direction rather than a precise gap. Most construction respondents plan to maintain their investment or raise it moderately, and almost none plan to cut it. In construction, the UK remains the most cautious region. There, 35 per cent of respondents plan no investment, down from 44 per cent, and only 5 per cent expect a significant increase. The report puts the minority with no plans to invest at around a quarter of each sector in one place and around a fifth in another, so treat its size as approximate. Measured commitment, not retreat, is the fair summary, with ambition running ahead of capability.
The RICS standard turns from what firms are doing to what they must do. Its title is Responsible use of artificial intelligence in surveying practice. RICS published the first edition in September 2025, and it took effect on 9 March 2026. It applies worldwide to RICS members and RICS-regulated firms. Where it conflicts with local law, the law takes precedence, and the firm must record the conflict and report it to RICS. Must marks a mandatory requirement, and should marks recommended best practice. Many duties apply to AI use with a material impact, meaning output that can influence the service delivered, and firms must record that judgement and the reasoning behind it.
The duties then divide by who carries them. The knowledge duty is personal. Any member using AI needs a basic understanding of how AI systems work and fail, including erroneous output, bias and the risks in how data is used. The governance duties fall mainly on firms. They need responsible use policies, a written register of the AI systems in material use, and a risk register covering bias, erroneous outputs and data retention, reviewed at least quarterly. A separate section sets requirements for firms that develop AI systems themselves, including a recorded assessment of their sustainability impact.
Now apply the standard to one pilot. Its requirements fall into five stages, which I will call decide, protect, inform, rely and review. The order matters, because the first three must be in place before the tool touches a client's documents. To decide, our consultancy records in writing its assessment of whether AI is the most appropriate way to review contract documents, and enters the tool in its register. To protect, it starts with the data. Client contract documents are confidential, so the firm obtains the client's express written consent before uploading them. With its supplier, the firm carries out due diligence through written requests, covering data protection, confidentiality, and any known gaps or bias in the supplier's data. Any question the supplier leaves unanswered goes into the risk register.
To inform, the firm tells the client in writing, before the tool is used, when AI will be used and for what purpose. Its terms of engagement explain how the client can contest that use, and whether and how the client can opt out. They also state the extent of any professional indemnity cover the firm holds for AI use. To rely, a named, appropriately qualified surveyor documents whether each material output, one that could influence the service, can reasonably be relied on, and if an output cannot be, the client is told in writing. This is the duty I opened with. On request, the firm must also give the client written information about the system and how its risks are managed. To review, the firm dip-samples outputs produced in volume at regular intervals, which means checking a random selection of them. It also reviews its risk register at least every quarter.
The rely stage leads straight back to skills, and the report's sharpest evidence on skills comes from free-text answers. One construction respondent was, in that respondent's words, concerned the younger newly qualified are already making use of AI, but implementing and blindly trusting it. Another wrote that AI could impact the ability to develop graduates. The two points connect. The tools are widely available at the stage of a career when experience is thinnest, so the problem is professional judgement as much as AI competence. The data fit this argument awkwardly. Skills are cited more often in construction, the sector further behind, while the report's reasoning would expect them to be felt most in the sector further ahead. In our consultancy, a graduate may run the document review, but a qualified, named surveyor must take responsibility for the decision to rely on it. The report also urges firms to set clear protocols and supervision for recently qualified staff. If you are early in your own career, this is where the standard reaches you first.
The survey also asked, for the first time, how far organisations align with the standard. Respondents answered when the standard had only just taken effect or was about to, so low alignment is no surprise. In construction, 43 per cent are not aligned, and most of them are simply unaware of the standard, 26 per cent of all respondents. In commercial property, 32 per cent are not aligned, and 19 per cent of respondents are unaware. In both sectors, then, more respondents are unaware of the standard than know of it and have not conformed, and partial alignment is the largest single group. The report concludes that the most direct route to alignment is reach rather than persuasion. That means bringing the standard into the channels surveyors already use for continuing professional development, or CPD. This alignment is also self-reported, and says nothing about how well any firm meets each requirement in practice.
The report ends with tasks shared between industry and RICS over three horizons. In the immediate period, 2026 to 2027, industry is asked to assess its AI use against the standard and to plan from the outset how a pilot will become routine practice. It should also document AI involvement in deliverables, including the tools, the data and the points of human review. RICS is to spread awareness of the standard, publish practical guidance and deliver CPD. In the medium term, 2028 to 2029, firms build the capability to convert pilots into production, with audit trails they can show to clients and regulators. RICS is to develop sector-specific extensions of the standard and bring AI competence into the Assessment of Professional Competence, or APC, the route by which candidates reach RICS membership. From 2030 onwards, industry scales what has proved itself, and RICS keeps the standard moving with the technology and the regulation around it. Bear in mind that RICS wrote both the survey and the standard, so its recommendations naturally centre on its own standard and programmes. Our consultancy's next move belongs to the immediate period, to plan the route from pilot to practice before the pilot ends.
That brings the evidence together, so return to the claim in the foreword. The survey shows that what has been crossed is the move from no use to piloting. Most organisations in both sectors now try AI, the pilot is the most common position, and embedded use remains rare. The foreword's claim holds for starting, then, and not yet for routine practice.
Quote each percentage from this report for what it is, the self-reported answer of a professional community close to RICS, gathered in 2026 and combined by weighting rules the report leaves unreconciled. Treat a pilot as unfinished until its route into routine work is planned, because the gap now lies in conversion, which skills, data quality and privacy slow. And where AI output informs a professional service, expect written decisions throughout. The one that matters most is the reliance decision, recorded by a qualified surveyor whose name is on it.
The report and the standard both reward reading in full. The episode on the NBS Digital Construction Report 2025 reads another industry survey with the same care. The episode on the RIBA AI Report 2026, from the Royal Institute of British Architects, turns to architects.
Reimagining AEC is written and recorded independently, and my name is Dr Sepehr Abrishami. Take the reliance question into the next AI pilot you join, run or study, at work or in research, and settle who will answer for its output before anyone uses it.