Goldman’s George Lee Sees a World Remade for Machine Intelligence

George C. Lee II has spent decades at Goldman Sachs watching technology reshape markets and economies. As co-head of the Goldman Sachs Global Institute, he now argues that artificial intelligence demands something more profound: a wholesale redesign of the physical and digital worlds built for human minds.

The pace of AI progress has already prompted bold forecasts. Elon Musk said this summer that AI may exceed the sum of all human intelligence in around five years. “There really won’t be anything that AI can’t do better than humans, apart from being human, perhaps,” he added. Anthropic CEO Dario Amodei envisions “a country of geniuses in a data center.” Lee takes these predictions as signals that the built environment must change.

Today our cities, offices, software interfaces and websites serve human cognition, work rhythms and trust. But what happens when machines operate at wire speeds with code-based contracts? Humans, Lee writes, may no longer serve as the primary actors. The Fortune article lays out this vision with striking clarity.

Physical spaces bend first.

History shows the pattern. Ancient Greek roads developed wheel ruts to guide carts. Railroads created logistical hubs that altered land values and commuting distances. The automobile produced multilane highways, vast parking lots and suburbs. Roughly 22 percent of land in large U.S. cities currently serves parking. Autonomous vehicles could slash that figure. Drop-offs replace parking. Merge lanes shrink because machines maintain tighter tolerances than anxious drivers. Roads narrow. Sidewalks widen.

These shifts have begun. In Dallas, Walmart stores cut openings in walls so Zipline delivery drones can load packages directly from the sales floor. The partnership with Uber announced in August accelerates drone use for consumer orders. Zipline co-founder Keller Cliffton noted that “every great transportation revolution has changed where people live, how businesses operate, and how economies grow.” New buildings may soon feature rooftop drone pads as standard.

Travis Kalanick, former Uber chief executive, pushes further. His Atoms robotics platform deploys specialized machines that make 1,000 pancakes an hour. Humanoid robots struggle with the same task. Kalanick’s CloudKitchens facilities, originally built for high-volume delivery prep, now retrofit for robotic workflows. “Digitizing the physical world is my life’s work,” he said. Utility trumps visual beauty. Simplicity and clear transit paths matter more than aesthetics.

Yet change arrives unevenly. Many structures remain expensive and slow to rebuild. Legacy human spaces will coexist with machine-native ones. Some human environments may become incidental. Others could turn sublime, freed from machine constraints. Capital might flow toward pure human experiences once machines handle routine operations.

Software faces parallel pressure. Large language models spawned AI agents that pursue goals with growing autonomy. Early agents mimicked clumsy human browsing. Now they chain APIs into complex workflows. Interfaces designed for human eyes become distractions. Agents need clean data surfaces, JSON schemas, service-level guarantees and telemetry.

This shift favors “headless” platforms. Aaron Levie, Box chief executive, observed that enterprises must ensure their software works across any set of agents. Dharmesh Shah of HubSpot has echoed similar themes. Systems of record persist as data repositories, but elegant dashboards lose primacy. Pricing may tilt toward usage or outcome models. The entire software value stack could compress.

The web transforms too. Cloudflare chief executive Mathew Prince reported in June that agentic traffic now exceeds human traffic online for the first time. Search engine optimization gives way to artificial engine optimization. Sites optimize for agents that browse, evaluate and transact on users’ behalf. Dynamic pricing, tokenized payments and machine-readable descriptions replace sliders, carts and drop-down menus.

Lee’s essay arrives at a moment of surging investment. Goldman Sachs Research forecasts global AI spending will exceed $1 trillion in 2026, with nearly $600 billion in the United States alone. A separate June report from Goldman Sachs Investment Banking, Harnessing AI for the Real Economy, argues the next boom lies in factories, utilities and industrial sites that represent 99.5 percent of global GDP. Software captured attention first. Physical AI comes next.

Recent warnings add nuance. On August 24, Goldman partner Chris Churchman told the firm’s Exchanges podcast that overreliance on AI risks “cognitive atrophy.” “There’s a huge danger here that in the era of AI, we outsource our reasoning to these models, and we have cognitive atrophy that stops us being able to reason from first principles ourselves,” he said. The CNBC report on his comments highlights the tension between productivity gains and skill preservation on trading floors and deal teams.

Earlier this year Goldman analysts examined the “world model” gap in current AI systems. Large language models excel at pattern completion but lack first-principles understanding of physics or causality. The April report, covered by Yahoo Tech, notes that solving this represents the next leap. Researchers including several AI pioneers race to build these models. Success would sharpen the very situational awareness Lee sees driving infrastructure redesign.

Job effects already surface. Goldman research released in August found AI-exposed industries show slower job growth since late 2022, especially in the United States. Call-center employment sits 39 percent below trend domestically. The Street report on the data suggests displacement pressures build faster than productivity statistics confirm. Yet corporate earnings calls still quantify AI impact in only 2 percent of cases, per related analysis.

Power constraints loom large. Goldman forecasts data-center electricity demand will rise 175 percent by 2030. Access to the grid, not just capital, will determine winners. Hyperscalers carry roughly $1.5 trillion in lease commitments. Utilization and revenue growth must keep pace. Recent X discussions among infrastructure investors highlight sustainability questions that remain open even as demand appears insatiable.

Lee’s piece improves on prior Goldman thinking. Earlier reports from the Global Institute, such as the 2023 essay on the generative world order, focused on productivity gains of 1.5 percent annually and $7 trillion in global GDP uplift over a decade. The current argument shifts emphasis from macroeconomic tailwinds to concrete changes in architecture, code and user experience. It moves beyond forecasts to vivid examples of walls cut for drones and kitchens rebuilt for pancake robots.

But, the transition carries friction. Enterprises must balance agent-friendly APIs with human oversight. Developers at Goldman itself now spend time “mentoring” AI tools with firm-specific knowledge, according to chief information officer Marco Argenti. Institutional memory must transfer into models without eroding human judgment. Churchman’s caution resonates here.

So the redesign unfolds on multiple fronts at once. Physical plants adapt for robots that ignore human form. Software sheds its graphical skin. The web speaks JSON to agents that outnumber human visitors. Capital reallocates. Land use changes. And humans? They may inherit spaces optimized for reflection or creativity once machines claim the cadence of wire-speed execution.

Lee does not present this as utopia or catastrophe. He presents it as logical consequence of machines that surpass human cognitive scale. The world that emerges will reflect the intelligence that shapes it. Whether that intelligence remains tethered to human values depends on choices made while the window for design still stays open.