How might we bring AI and autonomy into the 'built world' so that physical industries — construction, logistics, manufacturing — break through the productivity plateau?

The digital economy has seen explosive productivity growth over the past three decades: software scales instantly, information moves at light speed, and AI is now writing code, generating content, and automating knowledge work. But physical industries — construction, logistics, manufacturing, agriculture — have seen productivity growth stagnate or decline. Construction labor productivity is lower today than it was in 1970. Manufacturing employment has shrunk even as output stayed flat. Logistics costs as a percentage of GDP have barely budged despite decades of optimization. The 'built world' hasn't participated in the digital revolution because bits are fundamentally different from atoms.

Breaking through this productivity plateau requires bringing AI and autonomy directly into physical work: robots that can adapt to unstructured environments rather than repeating fixed motions, computer vision systems that can inspect quality in real-time across variable conditions, AI planners that can optimize logistics around real-world constraints, and digital twins that can simulate physical processes accurately enough to enable predictive optimization. The opportunity is enormous — physical industries represent over $10 trillion of U.S. GDP and employ 35 million people. Even small productivity gains would unlock hundreds of billions in economic value while addressing labor shortages, rising costs, and reshoring challenges.

The productivity gap between digital and physical industries emerged gradually over several decades, but its implications have become stark. In 1970, construction and software were both labor-intensive industries with similar productivity profiles. By 2020, software developers were 10-15 times more productive (measured by value created per worker) while construction productivity had actually declined. This divergence reflects a fundamental difference: software is malleable, copyable, and scalable, while physical work is constrained by atoms, energy, and the messy complexity of the real world.

Attempts to automate physical work have had limited success. Industrial robots, introduced in the 1960s, excelled at repetitive tasks in controlled environments like automotive assembly lines. By 2020, there were over 2.7 million industrial robots worldwide. But these systems are brittle: they require precisely positioned parts, fixed motions, and structured environments. They work in factories but fail in construction sites, warehouses, or farms where conditions vary unpredictably. The labor cost savings were often offset by the engineering cost of structuring the environment to suit the robots.

The AI revolution of the 2010s initially bypassed physical industries. Deep learning excelled at perception tasks — recognizing images, understanding speech, translating languages — but struggled with physical interaction. AlphaGo could beat the world champion at Go, but couldn't pick up the game pieces. GPT-4 could write essays, but couldn't pour a cup of coffee. The reason was partially technical (robotics hardware lagged AI software) but also economic: software companies could deploy AI improvements to millions of users instantly, while physical automation required custom hardware installations for each site.

The breakthrough came from combining several technologies: better sensors (especially depth cameras and LIDAR), more powerful and efficient AI models that could run on edge hardware, improved simulation tools for training robots in virtual environments before deploying physically, and new robot hardware with more degrees of freedom and better force control. By 2020, companies like Boston Dynamics were demonstrating robots with impressive mobility and dexterity, while startups like Covariant and Osaro were using AI to enable robots to manipulate objects they'd never seen before — a critical capability for real-world applications.

The COVID-19 pandemic accelerated interest in automation for physical work. Labor shortages, supply chain disruptions, and health concerns created urgent demand for solutions that could reduce human dependency in warehouses, factories, and logistics networks. Investment in robotics and automation companies surged. And as AI capabilities advanced, it became clear that the technology was finally ready to tackle unstructured physical environments. The question shifted from 'can we automate physical work?' to 'how quickly can we deploy automation at scale?' The answer will determine which countries and companies can reshore manufacturing, overcome labor constraints, and capture value in physical industries.