The Future of Construction: How AI is Transforming the Industry
Is construction ready for AI? Is AI ready for construction? I have no idea. There's a gigantic gap between people like me (technology optimists) who think AI will have many use cases in our outdated industry and the construction veterans hoping to make it to retirement before they need to adopt any AI tools.
Construction companies are sitting on a ton of data. Most companies I've talked to have a server that's basically a folder structure split up by projects, products, and employees. PDFs of construction plans, submittals, quotes. Separately there are emails, text messages, phone calls, and dozens of other sources of construction data.
Today, this data lives isolated and siloed. Need the submittal for the fans for the school project you're working on? Locate your projects folder, find the manufacturer, find the submittal, open it and review its approved status. Make a todo of the next task you need to complete.
In the future, this will be a single query to your AI platform: "is the submittal for the school project we're working on approved". The AI will use RAG to pull context from all your data sources and respond naturally:
"A submittal has been generated by Brian in your office, but it has not been approved by the engineer yet. You emailed the submittal to the contractor one week ago, would you like to follow up on the status?"
Connecting all these data sources creates a Jarvis-like experience. "Do we have enough money to purchase new computers for our employees this month?" Queries like this, with enough existing data, will be answered intelligently and faster than any employee could.
Generative AI will also disrupt how we create documents. Every document I just mentioned is typically manually created in construction. As a salesperson and mechanical engineer, I would manually create equipment schedules, budgets, submittals, quotations, and less obvious stuff like writing all my own emails. MEP firms do this on an even grander scale, manually generating (with the help of tools) complicated construction plans and specifications.
Many of these data sets will eventually be generated using niche construction models, like how LLMs generate language today. Ingesting millions of sets of mechanical drawings will help generative AI "predict" construction document variables and generate designs more quickly.
There's lots of low hanging fruit before we see generative construction plans:
- Taking emails from engineers and extracting data to auto-populate forms
- Using OCR on construction plans to generate material lists
- Summarizing project RFIs and change orders into digestible updates
AI will help with this type of generation first, then in the future help generate construction's more complicated models.
I've heard about the brain drain in construction as experienced professionals retire and aren't replaced by younger generations. This problem extends beyond knowledge workers to labor and installers as well.
I don't think AI will replace these humans in the coming years, but it can replace a lot of the lost "knowledge" through more easily queryable construction knowledge bases.
My first days in the industry were spent reading equipment installation manuals and learning how to interpret plans and specifications. I bothered many of my extremely patient colleagues for their insights constantly. I wish I could have asked AI questions like "What is the difference between hot gas reheat and hot gas bypass?" when those words didn't mean anything to me.
AI will help the newest employees make up that gap of knowledge lost as experienced construction workers retire. We'll have "Khan Academy" style AI bots for construction workers, from engineering students all the way through to retirees.
I expect adoption of AI in construction will be gradual, similar to full self-driving (FSD) technology in automotive:
| FSD Level | FSD Description | Construction AI Description |
|---|---|---|
| 0 | No automation | No AI assistance |
| 1 | Driver assistance for steering or acceleration/deceleration | Basic AI-assisted tasks |
| 2 | Partial automation - system has combined automated functions | Partial AI Assisted Systems |
| 3 | Conditional automation - can perform most driving tasks but human override required | AI handles most tasks, human oversight needed |
| 4 | High automation - fully autonomous in limited conditions | AI fully autonomous in specific conditions |
| 5 | Full automation - driverless operation in all conditions | AI fully autonomous in all conditions |
Construction AI will progress through these levels of autonomy. Currently, AI tools in construction are comparable to level 2 or level 3 FSD systems - assisting humans in specific tasks but not yet capable of complete autonomy.
It may take years or even decades to achieve "level 5 construction autonomy" where AI can handle all aspects of a construction project independently. Keep your hand on the wheel.