AI is moving at furious speed, but in the AEC sector, it’s running up against a brick wall in the form of outdated practices and legacy technology architectures that are notoriously slow to change, writes Richard Harpham, freelance industry commentator and fractional executive
Every industry eventually encounters a technology development moving far faster than the industry itself. In the retail sector, it was e-commerce. For the media, it was the smartphone.
In AEC, arguably one of the slowest moving professions in the developed world and one that is defined by longstanding habits and practices, it’s happening right now. AI tools multiply weekly. Every conference keynote presentation promises a step-change in productivity. And along the way, AI reinvents itself every few months.
At this year’s NXT BLD conference, I moderated a discussion with Amar Hanspal of Motif and Hugh McEvoy of Trimble. What struck me wasn’t the optimism expressed about what AI can build, but rather, how challenging it might be to charge for its usage (watch the panel discussion).
Discover what’s new in technology for architecture, engineering and construction — read the latest edition of AEC Magazine
👉 Subscribe FREE here
Outcome-based, deliverable-based pricing sounds like the logical future, or at least, once AI is capable of producing finished work, rather than just providing tools to streamline human-led working practices. As both panelists agreed, it’s brutally difficult to monetise AI in a way that keeps costs predictable for customers while simultaneously guaranteeing cash flow for vendors. That cash flow is necessary if vendors are to survive the lag before adoption, and thus revenue, finally catches up. It’s a lag that favours incumbents with strong balance sheets and quietly locks out start-ups that might otherwise be pushing the industry forward.
Walk into most architecture, engineering and construction firms today, and you’ll find something closer to anxiety than optimism about all of this. And that’s because the real pressure doesn’t derive from what AI can achieve. It comes from a much older problem that AI has simply made impossible to ignore any longer: namely, that this industry changes its processes, culture and even its commercial models at roughly one tenth of the speed at which the technology that underpins it is now moving. It’s in that disconnect, and not in the technology itself, that we see the real story emerge.
For decades, firms in the AEC sector have bought new tools repeatedly. But what gets neglected, is the underlying way that work gets done – who decides what, who checks what, who owns what, how information moves between one user and the next link in the chain.
AI is now emerging at a pace that the industry has never had to absorb before and is running up against a brick wall in the form of an unchanged technological foundation at many firms. The mismatch represents a bigger risk to most AEC technology strategies than any other risk that AI presents.
Four specific pressures show what this mismatch looks like in practice. Not one of them is uniquely tied to AI, but all of these pressures will get blamed on AI anyway, because that’s the easier narrative. And it’s here that the problem lies, because it allows the industry to avoid a harder conversation about why its own pace of change is the real constraint here.
No easy exit
The first pressure is that the Autodesk problem has no easy exit. Autodesk’s pricing has been climbing steadily since the company moved to subscription-only licensing, with list price increases compounding at roughly 6% to 9% a year and renewal discounts being quietly stripped away. For a firm running a few hundred AEC Collection seats, that’s not a rounding error. It’s a budget line that grows faster than fee income, year after year, with no ceiling in sight.
What makes this moment different from previous Autodesk price grumbling is that there’s nowhere obvious to go. Graphisoft is the only credible BIM alternative at scale, and for most firms, switching means re-platforming years of project data, retraining staff, rebuilding library content, and renegotiating workflows with every collaborator who isn’t also switching.
The firms moving fastest on AI adoption right now are not necessarily firms with the appropriate governance structures in place. That gulf is where the next generation of professional indemnity claims is going to come from and it will land on practices that genuinely believed they were being efficient, not reckless
And all that comes before you even get to the deeper problem: Graphisoft is itself part of Nemetschek, a consolidating competitor with its own commercial incentives, not a neutral open standard. Betting your practice’s core production tool on it isn’t obviously less risky than staying put. It’s a different flavour of walled garden, not an escape from it.
And the lock-in is getting more explicit, not less. Firms are starting to challenge Autodesk’s recent EULA terms that assert rights over what it terms ‘output’, the files and results that customers generate using its software, including restrictions on using that output to train AI models.
May Winfield, Buro Happold’s global director of commercial, legal and digital risks, has pointed out that this raises a genuinely uncomfortable question: if copyright law says the author owns what they create, but the software vendor’s contract tries to restrict how they can use it, which one governs?
Firms have spent decades treating their CAD and BIM output as their own intellectual property. That assumption is no longer as safe as it sounds, and most practices haven’t read the clause closely enough to notice. Winfield’s concerns extend well beyond IP: the same culture of insufficient scrutiny, she argues, is about to meet AI-generated deliverables, and the liability consequences will be harder to ignore.
This is the quiet crisis that lies beneath all the AI hype: a single vendor effectively controls the production layer for most of the industry, is raising the toll steadily, and is now reaching further into what customers are allowed to do with their own work. And there’s no credible replacement on the horizon. That’s not a software problem. That’s a structural one.
Culture problem
The second pressure is a culture problem. The clearest evidence of the mismatch in pace is what happens when AI lands inside a typical AEC workflow. Design review still happens in meetings and mark-up PDFs. Coordination still happens in scheduled clash-detection sessions, not continuous feedback loops. Information still moves between architect, engineer, contractor and supplier through formats and habits that predate the cloud, let alone generative AI.
None of this happens because better options don’t exist. It happens because of the industry’s culture, characterised by fragmented liability, fee structures that reward billable hours over outcomes and a deep institutional caution about anything that changes who’s responsible for what. In short, the AEC industry has never required anyone to change how they work, regardless of what tools arrived.
The result is that most of the AI productivity gains being celebrated right now are individual, not organisational. A designer generates renderings faster. An engineer drafts a calculation more quickly. A project manager summarises an RFI log in seconds, instead of an hour.
These are real gains and they’re worth having, but they’re also desktop-level gains layered on top of a structure that hasn’t shifted. The bottleneck in most AEC projects was never how fast one person could produce a drawing. It’s how slowly information moves between disciplines, how many times the same data gets re-entered into different systems, and how much coordination still happens through phone calls and PDFs, because the culture has never forced anything more structured.
Most of the AI productivity gains being celebrated right now are individual, not organisational, but the bottleneck in most AEC projects was never how fast one person could produce a drawing
Compare this to how other professions have absorbed comparable technology shocks. The legal sector didn’t just buy AI drafting tools. Firms restructured how associates are trained and billed once document review stopped requiring armies of juniors. In the same way, the financial services sector didn’t just buy algorithmic trading systems. The entire structure of how risk is priced and who’s accountable for it was rebuilt around the technology, repeatedly, and often under regulatory pressure.
AEC’s equivalent shift hasn’t happened yet, because the industry’s fragmented, project-by-project contractual structure diffuses responsibility for change across so many parties that no single actor is positioned to drive it. Until adoption is treated as an organisational and cultural project, and not a procurement decision, AI’s impact will stay capped at individual productivity – an improvement, certainly, but only a fraction of what’s possible.
Checking AI’s homework
The third pressure is the one that nobody wants to be first to admit applies to them. As AI-generated content moves from drafting drawings to writing specifications, generating calculations and producing reports, professional and legal exposure increases. Trusting output without proper verification is risky and yet most firms don’t have the governance to manage and mitigate that risk.
May Winfield of Buro Happold made this case directly to the NXT BLD 2026 audience: professional liability in AEC has always rested on the idea that a qualified person stands behind the deliverable and that a stamped drawing or a signed calculation means someone with professional responsibility checked it (watch her presentation).
AI doesn’t change that legal reality. It just makes it easier to quietly skip the checking, especially under deadline pressure and especially when the output looks plausible. The firms moving fastest on AI adoption right now are not necessarily firms with the appropriate governance structures in place. That gulf is where the next generation of professional indemnity claims is going to come from and it will land on practices that genuinely believed they were being efficient, not reckless.
This isn’t an argument against AI adoption. It’s simply an argument that proposes that adoption without verification discipline is a liability problem wearing a productivity costume. The same cultural inertia that slows process change also slows the development of the oversight habits that would make AI adoption genuinely safe.
Re-pricing human-written code
The fourth pressure is the quietest but possibly the most consequential for anyone building software in this space. Investors are increasingly marking down the value of human-written code as AI gets better at producing and replicating existing functionality. A codebase that took a team three years to build no longer commands the premium it once did, because a well-directed AI system can now approximate large parts of it in a fraction of the time.
This matters enormously for AEC tech specifically, because the sector is full of point solutions where the entire value proposition was “we built the hard, narrow thing”. If that hard, narrow thing is increasingly replicable, the moat was never the code. It has to be something else: proprietary data accumulated over years of real projects, switching costs baked into how teams work, or genuine network effects between the parties on a project. Software companies that can’t articulate which of those three things they own are going to find fundraising considerably harder over the next eighteen months, regardless of how good their product demo looks.
For an industry already nervous about Autodesk’s dominance, there’s a bitter irony here: the very things that made Autodesk hard to dislodge – decades of accumulated data, workflow lock-in, and network effects across the whole project ecosystem – represent exactly the kind of moat that AI-native challengers now need to build for themselves. Moreover, it’s exactly the kind of moat that pure code quality no longer buys.
And here, the monetisation problem bites hardest. A challenger that can’t charge reliably for outcomes, that offers a codebase that is no longer a defensible asset, and that is selling into a market that takes years to change its buying behaviour, isn’t facing just one of those problems. They’re facing all three at once.
Facing the consequences
As previously stated, none of these four pressures is fundamentally an AI story, even though AI will get the headline in most conversations about them. Autodesk’s pricing power and contractual reach is a market-concentration story made possible by a market populated by firms too slow-moving to credibly threaten to leave. Legacy workflows capping AI’s value is a culture story. Unverified AI output creating liability exposure is a governance story, made worse by adoption outrunning institutional caution rather than the other way round – for once. The repricing of human-written code is a capital-markets story about which industries have built real, defensible technology moats.
What ties them together is the same disconnect. Put simply, the pace at which AEC’s culture, contracts and ways of working evolve are radically out of sync with the pace at which the underlying technology is now moving, and that is the actual risk to every technology strategy in this sector – not AI capability, not vendor pricing, not any single point of failure.
Other comparable industries have been forced to compress decades of process change into a handful of years in order to keep pace with technology. AEC has not yet had that reckoning, and the only force that will bring it about is internal, with firms, software companies and clients all agreeing that adoption means changing how decisions get made and who is accountable for them, rather than which tool sits on the desktop.
Moderating a second NXT BLD session, this one on BIM 2.0, I saw the cost of that lag up close (see the panel discussion). Featuring founders from a new crop of BIM authoring solutions, this session surfaced that there may be a structural funding trap for start-up companies when trying to become the challengers the industry most needs. This is not a start-up execution problem; it’s a market structure problem.
These start-ups are serious companies building serious technology, founded between three and eight years ago. Yet it’s likely that most of them are not yet generating annual revenue higher than your favourite local restaurant. That’s not a comment on the quality of their products. It’s a comment on what happens when the industry that most needs disrupting is also the industry best designed to exhaust the people trying to disrupt it, entirely by habit and not by intention.
The new practices and start-up companies that emerge and succeed over the next few years won’t be the ones with the flashiest AI features. They’ll be the ones that treat the rate of process change as the issue that really needs fixing, because the technology is not going to slow down out of politeness.
Every other industry that has met a technology moving this fast has eventually been dragged, sometimes kicking and screaming, into matching its pace. AEC is no exception, whatever its habits suggest. The only real choice left now is whether that change happens on the industry’s own terms, deliberately and early, or whether it happens later, suddenly and to firms that just didn’t see it coming. The technology that won’t wait has never actually waited for anyone.
