Old Buildings, New Ideas

In the AI Story

AI tools currently function as old buildings in the digital economy. The comparison is precise, not metaphorical. A subscription to Claude Code costs roughly one hundred dollars per month at the individual level. For that price, a practitioner gains access to capability that, five years ago, would have required hiring a team of developers, purchasing enterprise software licenses, or spending months learning to code. The barrier to building has been lowered to the cost of a modest utility bill. The marketing manager who built her analytics dashboard did not need venture capital. The teacher who created educational tools did not need a grant. The architect who built structural analysis tools did not need an engineering background. Each found cheap space in which to experiment.

This is exactly the economic function Jacobs identified: old buildings make failure affordable, and affordable failure is the precondition for innovation, because innovation requires trying things that might not work, and no one tries things that might not work when the cost of failure is catastrophic. The explosion of AI-enabled building Segal documents in The Orange Pill is an explosion of experimentation made possible by cheap space.

But Jacobs would immediately ask a question the celebrants tend to skip: Who owns the building? An old building in a healthy city is typically owned by a local landlord — an individual or small company rooted in the neighborhood, with modest financial expectations. The AI tools that currently function as old buildings are not owned by neighborhood landlords. They are owned by some of the largest and most highly capitalized companies in the history of commerce. Their relationship to tenants is not the relationship of a neighborhood landlord to a local entrepreneur. It is the relationship of a real estate conglomerate to occupants of a development it can reprice, redesign, or demolish at any time.

The current pricing is not a permanent feature of the landscape. It is a business decision, made by companies investing far more in capability than they are recovering in subscription revenue, sustained by venture capital and corporate balance sheets optimizing for market share rather than profitability. When the investment phase ends and the monetization phase begins — when the companies that own the old buildings decide to charge rents that reflect the actual value of the space — the cheap space may cease to be cheap. The pattern has played out before, in both the physical and digital economies, with consistent results: the artists are displaced by luxury retail, the open web is absorbed by platforms, the bloggers migrate to feeds controlled by algorithms that serve advertisers rather than readers.

The question for AI tools is whether this cycle will repeat. The conditions for repetition are present: the tools are priced below cost, the pricing is sustained by investment capital rather than operating revenue, and the providers have structural incentives to increase extraction as they mature. The conditions for a different outcome are also present: open-source models provide alternatives, the cost of local inference is declining, and the community of builders is large enough to resist lock-in if alternatives exist.

Origin

Jacobs developed the argument in The Death and Life of Great American Cities (1961), specifically in her chapter titled "The need for aged buildings." She developed it further in The Economy of Cities (1969), connecting the physical economy of cheap space to the economic process of innovation. The AI-era application builds on her framework by reading subscription pricing as a form of rent and venture-subsidized tools as a form of cheap space with uncertain tenure.

Key Ideas

Cheap space is economic infrastructure. The availability of low-cost experimental space determines what enterprises can start.

New buildings cannot substitute. Their cost structure requires high-margin tenants, so they cannot host the low-margin experiments that produce innovation.

Ownership determines tenure. A neighborhood landlord and a real estate conglomerate are not functionally equivalent even when the current rent is the same.

Subsidized pricing is temporary by design. Venture-backed companies are optimizing for market share; the current rents reflect that strategy, not the underlying economics.

Open alternatives maintain the supply. Open-source AI models function as the continuously renewed stock of cheap space that prevents platform landlords from consolidating the market.

Debates & Critiques

Optimists argue that ongoing improvements in model efficiency, combined with the growth of the open-source ecosystem, will keep AI tools in the "cheap space" regime indefinitely — the rent may never need to rise because the underlying cost continues to fall. Skeptics note that the companies providing frontier models have no structural incentive to pass efficiency gains to users and that the historical pattern of platform monetization is nearly universal. The framework does not require predicting which scenario will prevail; it specifies that the conditions for continued experimentation depend on institutional choices rather than on the technology's trajectory.