How to Build a House in One Day

2023 April 14 Twitter Substack See all posts

Humanoid robots powered by artificial intelligence could build regular houses in one day using traditional construction techniques.

Rapid Construction Techniques

Building a House in Less Than Three Hours

Many people consider modern home-building techniques slow, but that is a misconception. One of my favorite videos is of a competition where workers build a standard stick-frame, code-compliant house in under three hours with no offsite prefabrication. There are more modern attempts, but the early 1980s vibes are way more fun. The pace at which builders construct modern houses has everything to do with economics and little to do with granular techniques.

Barriers to Everyday Adoption

Building a house in three hours requires an enormous planning effort. A team of professionals representing all trades took months to design the home and hammer out the construction strategy. A typical house build minimizes planning. Longer timelines are a side effect.

Another issue is labor slack. Many team members stand around and watch until the few minutes it takes to complete their assigned task. They would have little downtime using traditional schedules. It is also challenging to hire fifty plumbers at once!

Building Houses with Humanoid Robots

Humanoid robots could change the equation and unlock faster construction. Let's assume we have models like the proposed Tesla Optimus bot. They will have decent motor skills and artificial intelligence capable of independent problem-solving. Humanoid robots are beneficial because they can benefit from the tools and techniques we've spent centuries honing.

Rapid Context Switching Compresses Timelines

Homes under construction rarely have workers there. Workers take nights and weekends off, or they might be waiting on another contractor or inspector that can't fit immediately into the schedule. The key advantages of humanoid construction robots would be versatility and uptime. A single robot could run nearly 24/7 and have a carpenter program, a plumber program, etc. The robots can move immediately to the next task with no scheduling delays. Packs of robots can work on the house at once without the same downtime human crews with narrow specialties see.

It seems unlikely that it would be economical to build a house in under three hours. AI-powered robots may be better at coordination than humans, but there are still concerns about precision with so much work completed in parallel. Building things in series is fault-tolerant because workers can cut the next piece to adjust for variations in the previous work that warped lumber, the limited precision of mobile construction saws, a mistake, etc. might cause. Even stretching the timeline to a day would require 10x less work done in parallel than the sub-3 hour house. Each step in a series can be fast when you have dozens of robots working at once.

Another advantage of efficient context switching is reducing general contractor expenses. General contractors spend most of their day coordinating vendors and moving around/between sites. That work disappears under the flexible robot paradigm. They'll be robot wranglers instead.

A modern house needs between 1500 and 5000 man-hours, depending on its size and complexity. A team of one hundred robots could build most houses in a day!

Adapting the Supply Chain

Building a house in 24 hours requires an upgraded system for delivering materials. The easiest method is to have all the materials onsite (or along the street) at the start of the job. And it is probably a dead end to try to deliver the exact material requirements because miscounting, breakage, or defects could derail a work crew. A strategy I've seen work in the oil field is to have trailers full of generic materials. The supply company swaps the trailers out for restocking and charges for what workers use. Keeping 1.5 houses worth of supplies on location would only add twelve hours of inventory. The builder would still plan for custom items like fancy windows, plumbing finishes, etc.

Rising labor costs have pushed work offsite. Windows, doors, garage doors, and trusses are examples. Robot labor might bring some of that work back onsite. Could mixing concrete on location surge in popularity? The simpler the material inputs are, the easier it is to keep the crew supplied.

Cheaper labor could also increase the usage of alternative methods/materials like concrete walls.

Handling Bottlenecks

Some remaining bottlenecks are inspectors and customers.

The current inspection paradigm won't work, but it isn't all gloomy. The number of active construction sites would decrease by ~200x. You could probably afford to embed dedicated inspectors with each crew. A cheaper option might be to have a small team at City Hall that inspects houses by watching video streams from the robots. Any changes could happen in real-time as 100 robots swarm issues - no waiting for contractors to return.

Change orders can destroy any timeline. Many customers will stay onsite for the entire build to monitor progress. They will already pay for any delays they cause if construction companies are charging them time and materials. But under the robot paradigm, many more houses may be fixed-price bids. Builders might need a giant stopwatch that keeps track of the time the crew deviates from the plan on the customers' orders.

Adding Up the Financial Benefits

Tesla's cost goal for its robot is $20,000. It weighs ~75 kilograms, ~15x less than cars that cost the same amount. The cost could fall rapidly if manufacturing ever reaches scale. So it is conceivable that the total cost of robot laborers could be a few dollars per hour.

Roughly 60% of a new home's cost is construction, and 50% of the construction portion is labor. An 80% reduction in wages would reduce overall costs by 25%. If robots spread throughout the economy, trucking and materials prices could decline. Land costs might increase in such a world, but land-conserving multi-family structures use similar construction techniques as single-family homes.

The advantages of going fast are similar to the benefits of concepts like the Toyota Production System. The total inventory of uncompleted houses in the system declines, reducing carrying costs and making the system more responsive. There is no need for construction loans. Customers and builders won't have to bet on future economic conditions. Supply can satisfy demand quickly in jurisdictions that allow new construction.

Inventory also hides inefficiency and defects. If one project has issues, the workers can go to another without losing many working hours. It doesn't cost the general contractor responsible for the delay much money. Robot crews would pay much higher costs for going to alternative sites because idle robots hit the builder's bottom line directly. Any time spent switching locations is lost because robots work straight through without daily clean up, set up, and commute time like human workers have.

Builders would face immense pressure to keep the crew rolling. And humans often perceive things relatively. A building permit taking a few days or weeks appears acceptable when construction takes 6-12 months. But it seems absurd when a house gets built in a day. Cities will feel the heat to modernize their systems, especially after permit fees will likely grow to more than 10% of new home costs after other prices decline. A builder with an idle crew waiting on permits would spend all their time trying to be the squeaky wheel.

Direct costs should decline, but the effects of higher velocity will ripple through the whole system in beneficial ways.

The Return of Mega Projects

Mega projects would benefit from robot crews even more than home construction. Interest payments can be half the upfront cost of nuclear power plants that take 10+ years to build. Compressing construction time could halve the CAPEX.

The combination of rising labor costs and regulation kneecaps big projects in the rich world. Rising labor costs have pushed project managers to plan more precisely, modularize construction, parallelize work, and use more specialized workers and equipment. These systems are fragile to changes from regulators and other unexpected problems. The savings disappear with each change and require another overhead-heavy cycle of planning. An alternative is to return to onsite construction where sand, gravel, cement, rebar, and piping with few specialized components are the inputs for a swarm of robot laborers. Rapid context switching means more work can be done in series even as overall velocity increases dramatically. The combination of low planning overhead and low-cost labor handles hiccups much faster and cheaper than highly engineered, fragile construction paradigms. Nuclear plants will probably still be over budget, but the impacts should be less damaging. Less regulated projects like roads would fare even better (and cause less disruption for drivers). Factories can expand at a breakneck pace. Developers might build one gigawatt solar farms in a few days. Previously unfathomable projects seem plausible like millions of robots swarming to convert our high-voltage grid backbone from AC to DC in a few hours (increasing grid capacity 2x-3x).

Massive quantities of labor have their own quality, especially in construction. It is easy to be cynical about permitting, but plenty of places within the US have efficient permitting. The competitive pressures between jurisdictions will be extreme when a rural county on the Great Plains could become as grand as Dubai within a few months.

Building the Future

Robots with adequate motor skills and the software to power them are both challenging problems. It is hard to predict when we might solve them, but the recent advancements in artificial intelligence software and automotive technology breed optimism. Both technologies benefit from growing scale and investment, and the capital is available. A pessimistic case might look like self-driving software where demos are impressive, but everyday reliability is inadequate. A robot with decent motor skills might be too expensive or slower than a human. The counterarguments are:

  1. Construction bots can work with humans, while a self-driving car must have some autonomy.
  2. Cutting a board in the wrong place has fewer consequences than running over a pedestrian.
  3. Doing more work in series can limit rework.
  4. You can add more robots easier than trained humans.
  5. Some projects can afford to pay for speed.

It seems likely that crews would start with human handlers to provide feedback and point out errors. Labor bills might increase in the short term to pay human wranglers and larger robot crews, but many customers would pay a premium for rapid construction, driving the scale of robot production.

Artificial intelligence software alone might struggle to provide rapid productivity growth because only so many jobs can be 100% remote. Robots can augment the physical labor force, and demand will be insatiable if they can do jobs like construction. If a robot is half as physically capable as a human and can work 16 hours daily, the US will need ~11 million new robots yearly to grow the labor force at double-digit rates. The production rate will compound from there. The US produces ~10 million new cars per year, so the robots will be half the revenue of the car industry and ~1/15 the mass, assuming 75 kg $20,000 robots. Those are big numbers, but companies like Ford, GM, and Tesla have scaled production to these levels in a decade or two.

The future is bright, and I dream of the day a custom house is an impulse buy.