Wall Street Journal - Startup Targets New Frontier for AI: Construction Drawings

(Wall Street Journal) - New artificial intelligence tools are tackling a growing list of complex tasks: coding, drug discovery, financial analysis, legal reviews. Now add one more: reading construction drawings.

The startup Primepoint Labs has closed a $10 million seed round to advance an AI platform that interprets and connects construction documents. The round closed in two parts, the most recent in December, and was co-led by Navitas Capital, Penny Jar Capital and NextView Ventures. AI pioneer Yann LeCun, executive chairman at Advanced Machine Intelligence and former chief AI scientist at Meta Platforms, was among the investors participating.

“Until recently, AI couldn’t really move the needle in construction,” said Lubomir Bourdev, co-founder and chief executive of Primepoint. “Problems that construction is impeded by are things that AI couldn’t understand until recently, but LLMs [large language models] and technology that we’re building will start to make a big difference.”

Primepoint’s path

Bourdev and Hamid Palo co-founded the San Mateo-based company in 2024. Years earlier, Bourdev was a founding member of what is now known as Meta’s Fundamental AI Research, and LeCun was at one point his manager. Bourdev later co-founded WaveOne, a deep-learning video compression company that Apple acquired in 2023. Palo was an early employee at workflow-software company Trello.

After WaveOne, Bourdev was thinking about what to do next just as LLMs were taking off. He and Palo realized the potential for new AI tools to automate and coordinate work. “I started looking at construction and realized that it’s actually the most complicated collaboration,” he said.

Primepoint uses foundation models from AI companies like OpenAI and Google. But, it layers on its own algorithms to understand technical construction documents with complex instructions for sprawling teams. Projects currently using Primepoint’s software include a healthcare facility in San Francisco, a residential development in San Diego and student housing in Oregon.

In construction, project documents function as “the source of truth,” said Mike Heller, principal at Navitas Capital. And construction drawings are at the heart of them.

“There’s so much information in these things across thousands of pages,” Heller said. “And this information is really hard to access, but drives so many downstream workflows.”

Primepoint says its platform makes drawings interactive, saving time by centralizing information, linking data and flagging potential problems before a beam is installed or a wall is erected. Like AI startups in other fields, Primepoint believes its technology can automate tasks that now drain valuable time.

“I think the dirty little secret is—not all the time, but a good amount of time—the stuff just isn’t getting done, and that’s why you run into errors on the job site,” Heller said. People don’t have the time or desire to comb through thousands of PDF sheets to catch every little thing, he said, but AI is well-suited for that.

The ‘why now’ for construction

Investors are optimistic about AI’s potential in construction. Venture-capital deal value in U.S. construction tech last year was $1.54 billion, roughly double the amount in 2019, according to market-data firm PitchBook. AI is giving rise to a new crop of startups in the sector.

“The thing that excites me most is we, as a society and definitely as Americans, need to build,” said Joe Schmidt, a partner at Andreessen Horowitz, which isn’t an investor in Primepoint. New AI tools in construction are expanding productivity, helping to address the nation’s need for new data centers, infrastructure and housing, he said.

Schmidt co-wrote a recent blog post outlining AI’s potential for construction. “The entire built world is still running on software built in 1997,” the post said. “That’s about to change.”

Still, startups face challenges scaling in the construction industry.

A 2023 McKinsey and Co. article said, “… architecture, engineering and construction is one of the biggest industries in the world, but historically it has been among the slowest to digitize and innovate.” Startups looking to expand must sell to a large number of companies, meaning growth can be labor-intensive and slow, the article said, pointing out the average project involves more than 100 different suppliers and subcontractors.

Jose Luis Blanco, senior partner at McKinsey and global leader of the engineering, construction and building materials practice, said AI has improved efficiency and technology adoption in the industry, and could standardize data workflows and decision-making.

But the structural barriers remain. “AI hasn’t changed the fragmentation of the industry, the project-based nature of the procurement or even the complexity of decision-making,” he said.

“What AI has done is improving the ‘why now,’ but it’s still not the ‘why it’s hard,’” he added.