Construction has spent decades trying to digitize itself the same way every other industry did, by forcing everyone onto one universal data standard. It never worked. Now large language models are doing it differently, by reading the documents the industry already produces. Alain Waha, CTO of Buro Happold, calls it construction’s silicon moment, and he is not wrong.
For 30 years, the construction industry has been told it needs to digitize like everyone else. Banking did it. Manufacturing did it. Retail did it. Construction tried and failed, repeatedly, and the standard explanation has always been the same: the industry is too fragmented, too conservative, too resistant to change.
That explanation is wrong. Or at least, it is incomplete. The real reason construction never digitized is that the playbook every other industry used was structurally impossible to run here. Too many parties, too many standards, too many one-off projects. The industry could not get everyone speaking the same language because there was no single language to speak.
Then large language models showed up, and the problem solved itself. Not because anyone in construction built the solution. Because the technology, finally, was capable of meeting the industry where it already was.
Why every previous attempt at digitizing construction failed
The Tower of Babel problem nobody could solve
Every digitized industry has one thing construction never had: a shared data structure. Banks agreed on transaction formats. Retailers agreed on product codes. Manufacturers agreed on parts catalogs. Once an industry settles on a common language, computers can talk to each other, and digitization gets easy.
Construction has the opposite shape. About 90% of construction firms have fewer than 50 people. Every project is a temporary alliance of dozens of firms, trades, and specialists, each with their own software, standards, and ways of describing things. A door is a different data object in the architect’s model than it is in the contractor’s quantity takeoff than it is in the manufacturer’s spec sheet. Multiply that across thousands of components on a single job, and you start to see why construction earned its place as the second-least-digitized industry in the world, beaten only by agriculture.
The proposed solution, for decades, was always the same: get everyone on one universal digital standard. Open BIM. Industry Foundation Classes. National classification systems. Smart objects. The thinking was, if we can just convince every architect, engineer, contractor, and supplier on the planet to encode their information the same way, the data problem disappears.
How would you solve the Tower of Babel problem today? Would you try to get everybody to speak Esperanto, or would you connect them to Google Translate and make an LLM work?Alain Waha, CTO, Buro Happold
That is the question Waha keeps coming back to, and it lands hard. The industry spent 30 years trying to teach everyone Esperanto. The translator approach was never even on the table, because the technology to do it did not exist.
What changed in 2023
The technology finally meets the industry where it lives
Construction runs on documents. Always has. Drawings, PDFs, RFIs, submittals, contracts, specs, change orders, meeting minutes, daily reports. The information has always been there, it just was not in a form computers could do anything useful with. Until LLMs.
What an LLM does is read and understand language at scale. Not just text, but the kind of unstructured, messy, inconsistent language that construction documents are written in. A 200-page spec book with embedded tables and references to drawing sheets. A contract with clauses buried on page 187. A submittal that references three other submittals that reference five drawings. The thing that broke every previous attempt at digitization, the unstructured nature of the information, is exactly what LLMs are designed for.
That is the inversion. The old playbook said: change the industry to fit the technology. The new playbook says: build technology that fits the industry. And it is the first version of construction tech in 30 years where the smaller players can use the same tools as the largest. A 50-person GC running on Procore can use the same AI capabilities that a 3,000-person engineering firm like Buro Happold uses. That has never been true before.
- Document intelligence at the document level. Constructability reviews, RFI triage, submittal pre-checks, version comparison across thousands of pages.
- Cross-document reasoning. Catching the conflict between a spec, a drawing, and a schedule before it becomes a change order.
- Knowledge that scales. Junior engineers asking complex project questions and getting cited answers grounded in the actual project documents.
Real money is already moving on this thesis. Primepoint just raised $10M to build AI that reads construction drawings, with Yann LeCun on the cap table. Andreessen Horowitz published a thesis arguing the design software stack is overdue for a rebuild, and we covered why that matters for contractors and software vendors. The pattern across both: targeted document intelligence, not enterprise wrappers.
The construction briefing built for people who actually run projects
News, deals, and deep takes on the forces reshaping construction, written for people who have to make decisions on Monday morning.
Join 2500+ ReadersSteel, electricity, silicon
Why Waha thinks engineers are blessed, not doomed
Most AI commentary right now is about replacement. White-collar jobs going away, SaaS revenue collapsing, the end of work as we know it. Sit down with the CTO of a 3,000-person engineering firm and the framing flips entirely.
We are blessed. We have been given a new concrete, a new steel. Last time the built environment really changed was Brunel. We got steel and we could go up. We got electricity and we could light our spaces. We now have silicon.Alain Waha, CTO, Buro Happold
The framing matters because construction is not a sector where AI threatens to wipe out the workforce. The industry is structurally short of people. The US needs roughly 439,000 additional workers to meet current demand. The UK construction workforce has hit a 25-year low. London alone has a million-home shortfall. Net zero targets are tightening, regulation is tightening, and most projects still finish late and over budget. These are not problems AI makes worse. These are problems where the industry has been waiting for a tool capable of handling the complexity, and one finally exists.
That is the WIIFM for anyone running a construction business right now. AI in construction is not about doing the same work with fewer people. It is about being able to take on the work the labor shortage forced firms to turn down. McKinsey tracks $50 billion in venture and PE capital that flowed into AEC tech between 2020 and 2022, and the median deal size has doubled since 2017. The capital flow alone tells you investors are betting this is the moment.
What this means in practice
How to think about it as an operator vs. a founder
For construction executives, the practical question is not whether to adopt AI. It is which use cases pay back fastest in your specific operating model. The strongest early signals are in document-heavy, high-rework workflows, where the cost of missing something is meaningful and the language patterns repeat. Constructability review, contract risk extraction, RFI drafting, submittal triage. Not coincidentally, those are the categories where most of the recent ConTech funding is concentrated.
For founders building in this space, the bigger lesson from Waha’s framing is that the technology has finally caught up to the industry, not the other way around. That changes the build philosophy. Stop trying to make construction operate like software. Build software that operates like construction already does. The companies winning right now are the ones that ingest the actual document set, build a knowledge graph, and let humans verify the AI’s first pass.
| Approach | Old digitization playbook | LLM-era playbook |
|---|---|---|
| Core idea | Standardize the industry | Read the industry as it is |
| Required input | Structured data, tagged objects | Existing documents in any format |
| Adoption barrier | Every party must change tools | Works on top of current workflows |
| Who can use it | Mostly enterprise firms with budget | Same tools accessible to small contractors |
| Time to value | Years, often a decade | Months, sometimes weeks |
Frequently asked questions
The approach assumed every party in a construction project would adopt the same data standard, encode information the same way, and maintain it across the project lifecycle. In a fragmented industry where about 90% of firms have fewer than 50 people and every project is a temporary alliance, getting universal compliance was structurally impossible. McKinsey’s research found construction productivity grew just 1% per year over two decades, against 2.8% for the global economy.
The strongest current use cases are constructability reviews, contract risk extraction, RFI drafting, submittal pre-checks, and version comparison across thousands of pages of drawings and specs. We have covered the document review category in detail here. The common pattern is targeted document intelligence rather than broad enterprise AI.
Construction is structurally short of people, not long. The US has a roughly 439,000-worker shortfall, the UK workforce is at a 25-year low, and London alone needs a million more homes. Unlike SaaS or services, construction still has to physically build things. The bottleneck is capacity, not headcount, which is why most operators frame AI as a capacity unlock rather than a replacement tool.
Two things changed simultaneously. LLMs and vision models can now parse the unstructured documents construction actually runs on, which previous systems could not. And the same tools are accessible to small and large firms alike, removing the historical pattern where only enterprise contractors could afford serious tech. McKinsey’s UK blog tracks $50 billion of capital flowing into AEC tech between 2020 and 2022, with median deal size doubling since 2017.
Start with one document-heavy workflow where the cost of missing something is high and the language patterns repeat. Contract review, constructability checks, and submittal triage are the strongest early use cases. Pick a tool that connects to your existing project management stack rather than asking your team to switch systems. The firms moving fastest are the ones treating this as workflow automation, not enterprise transformation.
Primepoint Raises $10M Seed to Build AI That Actually Reads Construction Drawings – Bricks & Bytes
The VC Firm Behind Facebook Just Told Construction: Your Software Is Next – Bricks & Bytes
Reinventing Construction Through a Productivity Revolution – McKinsey Global Institute
How the Construction Industry Can Boost Productivity Through Technology – McKinsey UK
Alain Waha, Chief Technology Officer – Buro Happold
Related Articles
How AI Is Automating Construction Document Review
Primepoint Raises $10M Seed to Build AI That Actually Reads Construction Drawings
The VC Firm Behind Facebook Just Told Construction: Your Software Is Next
