From the building site to the factory
John Werner, a Forbes author, took part in a panel at the Minds in Motion event at MIT’s Samberg Conference Center in Cambridge. It took place a day before Halloween. Werner is affiliated with Imagination in Action, the event’s organizer, helped put together the program, and is also affiliated with Link Ventures.
The panel was moderated by Fraser Anderson of Link Ventures. The participants included Vikas Enti of Reframe Systems, Michael Santora of Logic, and Mauricio Pereira of Industrial Automata.
The discussion covered automated factories, autonomous building sites, and supply chains. Robots can already perform a substantial share of construction work, while AI agents can help design and manufacture components collaboratively. Robots for nailing, carpentry, electrical work, and plumbing are nearing deployment. Participants discussed which tasks will remain for people in 10, 15, or 30 years, and whether buildings will be designed and constructed almost entirely autonomously.
Enti said automation should start with evaluating the process, not an individual robot. He described construction as a linear, fragmented, poorly organized process: automating individual tasks within the current setup does not deliver the greatest benefit. Reframe Systems breaks the process into parts and moves it into a controlled environment: a factory.
There, a single robot can work on several tasks at once, spend less time idle, and deliver more value. The company had to develop a production architecture and test the economics of moving construction into a factory, including whether the business model could work. Reframe Systems initially placed particular emphasis on robotics.
The company built a “pixels to parts” technology stack for mass customization. Enti said this makes it possible to turn construction into a manufacturing process while still responding to market demand. Traditional industrial manipulators and modern computer vision systems mean robots do not have to be programmed to repeat one task endlessly; they can switch between different jobs:
End-to-end supply chains and data
Santora spoke about the complexity and fragmentation that AI systems face. Logic’s robots, he said, can move between multiple sites regardless of who owns or operates them. In construction, materials have to be delivered from production to a site, placed in storage, and then transported onward — by truck, train, or other means.
Santora argued that each stage should use a single robotic system with a shared control layer, rather than a set of disconnected solutions for separate tasks. Fragmented automation adds complexity; sometimes it is simpler to do the work by hand.
A unified system can track who owns the cargo and collect data along the entire route, from production through assembly, acceptance, or another final process. Santora said that data could help train the AI models that will manage these industries in the future.
Construction costs and timelines
Pereira discussed expenses at building sites. By his estimate, of every 100 dollars spent on labor, 44 dollars go toward insurance and liability. If robots can do the same work, they could reduce costs that currently produce no direct construction output.
Pereira also linked automation to lower housing costs. A controlled environment, he said, is necessary for applying physical AI in manufacturing. Robotic foundation models can already be used on site after fine-tuning and skill development, handling the challenges that arise outside a factory.
Enti suggested that future generations will be surprised that buildings were constructed directly on site. Under his scenario, buildings that now take years could be manufactured and installed in a few days — almost instantly by today’s standards.
Long construction timelines affect the housing market: while homes are being built, demand changes, leading to housing surpluses and shortages. Enti proposed an alternative to speculative construction: guarantee customers a home a week or a month after they place an order. That would require a predictable, highly parallelized process not constrained by the physical conditions of a building site. In his view, a factory makes such a process possible.
Construction and logistics in space
Santora discussed using space beyond Earth’s stratosphere for cargo storage and future construction. He believes space will accelerate the development of autonomy in warehousing, supply chains, and construction, going beyond simple automation. His company is involved in an undisclosed project related to deep space.
The project aims to move cargo from Earth into space as part of a single system. Prepositioning cargo should reduce costs. Santora estimated that delivering one kilogram to a destination on the Moon costs about 4.4 million dollars: around 1.5 million dollars to reach the general destination area, plus another 2.9 million dollars to reach the landing site. If the cargo is not ready for use there, human intervention in space could cost millions of dollars. A connected supply chain that delivers goods from the commercial supply chain to the required location without additional cargo handling should, in Santora’s view, bring costs down as far as possible.
He also gave an example of how transportation history constrains modern systems. Ruts left by Roman carts forced later carts to follow the same route. Cartwheel manufacturers later made wheels for trains, and trains now carry rocket parts. Santora argued that space logistics must take earlier decisions into account without carrying over practices that are harmful on Earth.
Projects and autonomous robots
Pereira spoke about a future in which 10,000 launches a year are expected — far more than today. He suggested that physical AI agents will emerge, similar to existing agents that create software and cycles. Large language models or programs could check projects against building codes and safety requirements, while autonomous systems could construct buildings. Physical AI could ask users questions, and agents could create software.
In Pereira’s scenario, autonomous robots would make it possible to build buildings and change their appearance as desired, lowering the technical barrier to entry. Anyone could subscribe to a system that designs a building according to the principles used by a structural engineer, then ask robots to manufacture a safe structure that meets code.
Anderson added that, thanks to “superintelligences,” these systems would process more data in real time at the edge. He contrasted this with the idea of automation as the mechanical execution of tasks that strips work of its human dimension: he said large language models and the superintelligence being created are not mechanistic. Anderson then ended the discussion. Werner remarked that the conversation offered plenty to think about when it comes to the future of homebuilding.
Dataist’s take: Factory assembly could make construction timelines more predictable, but the participants’ scenarios still depend on how well design, production, delivery, and robot operations can be connected. Even predictions of construction in days assume not only advances in machines, but a redesign of the entire process.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X