The emergence of new technology has long inspired humans to shape our environments. The invention of steel permitted the rise of skyscrapers, the automobile inspired highway systems and the internet gave rise to a whole new set of digital spaces.
Now, a new force for change is upon us. The pace of improvement in AI models has led many observers to predict that human intelligence may soon be overtaken by the cognitive capacity of machines. Anthropic CEO Dario Amodei imagines a world with “a country of geniuses in a data center.” Elon Musk posited this summer that “AI may exceed the sum of 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.”
In such a world, it seems likely that the digital and physical affordances we have constructed for the benefit of humans will be superseded by or at least complemented by infrastructure that is specifically designed for ready and effective use by “bots,” both digital and physical.
Today, our world is designed for human cognition, cadence and trust. Now imagine a world built for machine intelligence, wire speeds and code-based contracts. Humans may no longer be the primary actors in this new theater of our own making.
Bending and Breaking Physical Spaces
AI agents may transform the physical world. It seems superfluous to say that our homes, offices, stores, and public spaces are designed for humans. Think of the centuries of architectural design dedicated to making those spaces useful, accommodating, safe, and efficient. Never mind the time and creativity devoted to making them aesthetically pleasing. But as robots begin to populate our society and workplaces, these locations will need to be accessible and effective for humans and robots alike. Both will need architecture and design for human-machine cooperation.
Building new physical structures is often a slow and expensive process. But humans have rebuilt our physical world to accommodate new technologies in the past, with changes in transportation offering perhaps the clearest examples. Ancient Greek roads had wheel ruts to guide carts over steep or slippery terrain. Railroads necessitated new logistical hubs and services around rail terminals. This had downstream effects, changing the value of land, expanding the potential to commute long distances to cities, and creating physical dividing lines between neighborhoods.
In the 20th century, the rise of mass-produced automobiles led to multilane highways, the construction of new parking facilities, and the rise of the suburbs. In our own time, how might autonomous vehicles reshape the physical environment? Autonomous vehicles do not have the same requirements as traditional cars with human drivers. A human driver requires parking near the driver’s end destination. An autonomous vehicle could drop off a passenger and move on to its next task. Alternatively, it could queue elsewhere, waiting for pickup time. Today, roughly 22% of land in cities with over 1 million inhabitants is used for parking. The rise of autonomous vehicles could reduce that figure, opening land to other uses. Merge lanes could be smaller, given that AVs often operate with tighter tolerances for risk than human drivers. Roads could also be narrower, with expanded room for sidewalks or bike lanes. Furthermore, without the need for steering wheels or pedals, the physical shape of cars may change, opening the possibility for new forms for the cars of the future.
Such changes won’t happen overnight, but our physical world is already being rebuilt to accommodate AI-enabled technologies. In Dallas, Texas, efforts are underway to prepare for the advent of Zipline’s drone-based delivery service. Born from a successful effort to speed the delivery of blood plasma to remote hospitals in Rwanda, Zipline has evolved to serve consumers with fixed-wing drone deliveries of everything from Starbucks to DoorDash. One of Zipline’s primary partners in the region is Walmart, which is beginning to adapt its store design to drive more convenience and efficiency in staging deliveries by drone.
Workers at one such facility recently began cutting openings into the walls of a Walmart Supercenter — not for a renovation, but for delivery drones, letting employees load packages directly from the sales floor into the sky, bypassing the free-standing charging posts the companies used at first. It’s a small, literal crack in a retail architecture built for nearly a century around one assumption: that the customer walking through the front door is human.
The impact of such efforts could soon spread far beyond individual facilities. In August Zipline announced a partnership with Uber to accelerate the use of drones in deliveries. Noting the potential effect of such parentships beyond how quickly customers could now get their next order, Zipline co-founder Keller Cliffton stated, “Every great transportation revolution has changed where people live, how businesses operate, and how economies grow.”
Ultimately, supply stations for Zipline may become purpose-built for these kinds of workflows. New apartment buildings may be designed with convenient landing spots for the drones to drop their payloads on rooftops or courtyards.
While many efforts exist to build humanoid robots that share our form and therefore can operate reliably in our environments, it is likely that such machines will be outnumbered by robots that abandon the human form in favor of utilitarian shapes designed for efficiency. As former Uber CEO Travis Kalanick observed in the public unveiling of his new “Atoms” platform, a specialized robot that looks nothing like a human chef could optimally tackle the task of “making 1,000 pancakes an hour,” while a humanoid, not optimized for that function, would struggle to complete this enormous task.
Some transformations are even further along. Kalanick’s CloudKitchens business, a subsidiary of Atoms, is also shifting further in the direction of purpose-built infrastructure for robots. With the remarkable rise in online food delivery, many restaurants struggle to balance service and food prep for diners in their establishments while also preparing staging and queuing delivery for takeout orders. Kalanick initially addressed that challenge by offering restaurateurs new spaces that were solely for the preparation of meals for delivery and were located strategically around cities and suburbs. Supplies came in on one side via loading docks. Kitchens were structured for high throughput and delivery vehicles lined up on the other side of the building to take finished meals and speed them along critical thoroughfares for timely and low-cost delivery. Now, that model is shifting to retrofitting these facilities for robotic food prep and kitchen design optimized for robots rather than chefs. With these new technologies, as Kalanick says, “digitizing the physical world is my life’s work.”
As we reconfigure our physical spaces for robotic collaborators, expect new expectations and disruptions. Humans want physical infrastructure built for their comfort and efficiency, with an eye towards aesthetic beauty, or at least the familiar; robots function best with simplicity, easy transit and “beauty” expressed in utility not visual appearance.
The End of Software (As We Know It)?
The software ecosystem is already being re-shaped for the convenience of AI agents. AI agents, born from large language models (LLMs), are designed to pursue goals on behalf of users with some degrees of autonomy and reasoning capability. They also feature the ability to invoke and utilize computing resources such as browsers, websites, applications and data stores to help accomplish these goals.
Progress in this area has been rapid. The state of the art has quickly evolved from early demonstrations of agents navigating web pages to perform online shopping, a mode that visually approximates your grandparents learning how to use Amazon.com in 1999, tentatively clicking around, back-spacing and often invoking the wrong commands. Now, agents are capable of performing sophisticated workflows and traversing multiple applications by leveraging a critical artifact of modern computing, the application programming interface (API).
APIs have become a valuable and dominant way to connect applications to data sources and to each other. Think of them as on-ramps and off-ramps that connect highways to cities—and other highways. They were part of the arcane plumbing that lies beneath the foundation of our increasingly well-crafted software and workflows. Humans engaged user interfaces to command software and engage with outputs like dashboards, while APIs labored in the background.
Now, AI agents can command individual applications and autonomously compose those API interactions into multi-step workflows. In this world, APIs are not only the plumbing, they are the interface. Agents don’t require a beautiful canvas to function. Rather, they need to access applications seamlessly, go right to the heart of the data store or logic layer to perform an operation and then move on to the next step. Agents favor applications that are set up to help them navigate – modern APIs, machine-readable content that functions as a user guide to the platform, service-level guarantees, transaction capability and telemetry to assess effectiveness.
These shifting requirements signal the rise of “headless software,” platforms that are optimized for agents and that favor utilitarian interfaces over elegant design. This evolution has been well described by leading software entrepreneurs such as Dharmesh Shah, co-founder and CTO of HubSpot, and Aaron Levie, founder and CEO of Box. Levie has observed that “enterprises need to be able to ensure all of their software works across any set of agents they choose.”
With the rise of headless software platforms, SaaS applications act increasingly like data repositories that agents can traverse and stitch into complete workflows. To the extent this paradigm continues to emerge, it implies a kind of relegation of some classical software applications with their elegant interfaces designed to engage humans. So-called systems of record remain valuable as reliable, persistent stores of corporate data, but they become a watering hole along the agentic journey, not a destination. Consequently, the value proposition of classical (in other words, human-centric) software may shrink commensurate with this new role and its pricing power. Perhaps the entire pricing model shifts in favor of usage or outcome-based revenue models that better align software vendors with customers’ desire for value realization.
The Worldwide Agent Web
Websites are undergoing a similar transformation. Decades of work to perfect the human appeal of websites’ user interfaces and commerce sites allow us to browse and shop in a familiar manner, loading virtual shopping items into virtual carts, and seeking our own optimal combination of price, quality and availability. However, such designs may now be superseded by austere sites that allow agents to act efficiently on our behalf.
Each form optimizes for different functions and users. Humans browse the internet looking for images, drop-down menus, buttons, slider bars, folders, dashboards, and web forms that allow us to interact with computing in ways that are familiar and intuitive. But to AI agents, these features are distractions, barriers to their direct access to data, capabilities and logic they need to complete tasks. Humans want intuitive interfaces and appealing visuals. AI agents want “clean” API surfaces, a descriptive markdown file, and JSON schemas.
The transformation of the internet to more agentic interfaces is already underway. As Mathew Prince, the co-founder and CEO of Cloudflare, a leading internet infrastructure company, Cloudflare, said in June of this year, “Agentic traffic [is] growing so fast that bots have now passed human traffic online for the first time in the Internet’s history.”
As agents traverse the web on our behalf, the practice of Search Engine Optimization (SEO), through which websites compete for human visitation by tuning their appeal and seeking referral of users from Google and other discovery platforms, is giving way to Artificial Engine Optimization (AEO), which is designed to induce agents to promote, visit and even transact on websites by increasing their visibility and appeal to our digital delegates.
As with software, the features that optimize for agents differ significantly from the intricate features of modern internet sites built for human use. Commerce and content purveyors are scrambling to contribute their data to train LLMs and place themselves squarely in the transactional path of agents. The implications for online commerce are particularly significant: The parallel processing capabilities of AI can allow agents to comparison shop at scales and speeds far beyond the capacities of humans.
Millions of websites have advantages of legacy and incumbency. But over time, people’s loyalty to online brands may be supplanted by the efficacy of their interfaces for agents. Unlike humans, agents don’t shop habitually or “get used to” shopping on any given platform. Dynamic pricing may become more pervasive amid these accelerated shopping sprees. A logical outcome for this development could be the rise of auction pricing at scale, where agents put out “requests for proposals” for every sweater or light bulb purchase they make on our behalf and induce online stores to compete to win every piece of business in real time. Tokenized payments may rise in use for agent-based transactions, where immutability and speed are most desirable. By the same token, tolling infrastructure may emerge that permits agents to autonomously “pay for” access to content or other online resources as they pursue our goals. Both of these possibilities could meaningfully challenge the current economic structure of the web, where human-driven search lies at the heart of online monetization and existing payment methods and rails are necessary to consummate transactions.
How this Might Play Out
We are beginning to see evidence that our digital and physical spaces are evolving with the advances of AI and robotics. This raises a series of critical questions. How dominant will “machine spaces” become? What models for co-existence and control will emerge? Will “human spaces” become a rounding error too?
Many of these changes are nascent. As such, they will coexist with established architectures. In these early years, we should expect the development of “parallel universes.” Software and websites built for humans won’t vanish. Factories, warehouses and stores won’t suddenly close. But a new and different infrastructure will eventually emerge – one that is designed to serve our digital workers in contexts where humans may no longer be the primary actors.
We may begin to consider the ergonomics of agents and robots as much or more than we consider optimization for humans. While some may view this wistfully, it seems likely that these changes and the tremendous efficiency gains they promise may free up capital and creativity to build new, entirely human-centric architectures that are not burdened by the compromises necessary to accommodate people and machines but rather are tuned exclusively to our highest tastes and aspirations.
The irony is that a world redesigned for machines may eventually allow us to recover something more purely human. If factories, kitchens, warehouses, websites, and workflows become increasingly machine-native, then human spaces may be relieved of some of their utilitarian burden. We may build more places for beauty, reflection, play, learning, and community precisely because the machine world has absorbed more of the work. The danger is that human spaces become incidental. The opportunity is that they become sublime.
Goldman Sachs Global Institute co-head: remaking our world for machine intelligence
The emergence of new technology has long inspired humans to shape our environments. The invention of steel permitted the rise of skyscrapers, the automobile inspired highway systems and the internet gave rise to a whole new set of digital spaces.
Now, a new force for change is upon us. The pace of improvement in AI models has led many observers to predict that human intelligence may soon be overtaken by the cognitive capacity of machines. Anthropic CEO Dario Amodei imagines a world with “a country of geniuses in a data center.” Elon Musk posited this summer that “AI may exceed the sum of 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.”
In such a world, it seems likely that the digital and physical affordances we have constructed for the benefit of humans will be superseded by or at least complemented by infrastructure that is specifically designed for ready and effective use by “bots,” both digital and physical.
Today, our world is designed for human cognition, cadence and trust. Now imagine a world built for machine intelligence, wire speeds and code-based contracts. Humans may no longer be the primary actors in this new theater of our own making.
Bending and Breaking Physical Spaces
AI agents may transform the physical world. It seems superfluous to say that our homes, offices, stores, and public spaces are designed for humans. Think of the centuries of architectural design dedicated to making those spaces useful, accommodating, safe, and efficient. Never mind the time and creativity devoted to making them aesthetically pleasing. But as robots begin to populate our society and workplaces, these locations will need to be accessible and effective for humans and robots alike. Both will need architecture and design for human-machine cooperation.
Building new physical structures is often a slow and expensive process. But humans have rebuilt our physical world to accommodate new technologies in the past, with changes in transportation offering perhaps the clearest examples. Ancient Greek roads had wheel ruts to guide carts over steep or slippery terrain. Railroads necessitated new logistical hubs and services around rail terminals. This had downstream effects, changing the value of land, expanding the potential to commute long distances to cities, and creating physical dividing lines between neighborhoods.
In the 20th century, the rise of mass-produced automobiles led to multilane highways, the construction of new parking facilities, and the rise of the suburbs. In our own time, how might autonomous vehicles reshape the physical environment? Autonomous vehicles do not have the same requirements as traditional cars with human drivers. A human driver requires parking near the driver’s end destination. An autonomous vehicle could drop off a passenger and move on to its next task. Alternatively, it could queue elsewhere, waiting for pickup time. Today, roughly 22% of land in cities with over 1 million inhabitants is used for parking. The rise of autonomous vehicles could reduce that figure, opening land to other uses. Merge lanes could be smaller, given that AVs often operate with tighter tolerances for risk than human drivers. Roads could also be narrower, with expanded room for sidewalks or bike lanes. Furthermore, without the need for steering wheels or pedals, the physical shape of cars may change, opening the possibility for new forms for the cars of the future.
Such changes won’t happen overnight, but our physical world is already being rebuilt to accommodate AI-enabled technologies. In Dallas, Texas, efforts are underway to prepare for the advent of Zipline’s drone-based delivery service. Born from a successful effort to speed the delivery of blood plasma to remote hospitals in Rwanda, Zipline has evolved to serve consumers with fixed-wing drone deliveries of everything from Starbucks to DoorDash. One of Zipline’s primary partners in the region is Walmart, which is beginning to adapt its store design to drive more convenience and efficiency in staging deliveries by drone.
Workers at one such facility recently began cutting openings into the walls of a Walmart Supercenter — not for a renovation, but for delivery drones, letting employees load packages directly from the sales floor into the sky, bypassing the free-standing charging posts the companies used at first. It’s a small, literal crack in a retail architecture built for nearly a century around one assumption: that the customer walking through the front door is human.
The impact of such efforts could soon spread far beyond individual facilities. In August Zipline announced a partnership with Uber to accelerate the use of drones in deliveries. Noting the potential effect of such parentships beyond how quickly customers could now get their next order, Zipline co-founder Keller Cliffton stated, “Every great transportation revolution has changed where people live, how businesses operate, and how economies grow.”
Ultimately, supply stations for Zipline may become purpose-built for these kinds of workflows. New apartment buildings may be designed with convenient landing spots for the drones to drop their payloads on rooftops or courtyards.
While many efforts exist to build humanoid robots that share our form and therefore can operate reliably in our environments, it is likely that such machines will be outnumbered by robots that abandon the human form in favor of utilitarian shapes designed for efficiency. As former Uber CEO Travis Kalanick observed in the public unveiling of his new “Atoms” platform, a specialized robot that looks nothing like a human chef could optimally tackle the task of “making 1,000 pancakes an hour,” while a humanoid, not optimized for that function, would struggle to complete this enormous task.
Some transformations are even further along. Kalanick’s CloudKitchens business, a subsidiary of Atoms, is also shifting further in the direction of purpose-built infrastructure for robots. With the remarkable rise in online food delivery, many restaurants struggle to balance service and food prep for diners in their establishments while also preparing staging and queuing delivery for takeout orders. Kalanick initially addressed that challenge by offering restaurateurs new spaces that were solely for the preparation of meals for delivery and were located strategically around cities and suburbs. Supplies came in on one side via loading docks. Kitchens were structured for high throughput and delivery vehicles lined up on the other side of the building to take finished meals and speed them along critical thoroughfares for timely and low-cost delivery. Now, that model is shifting to retrofitting these facilities for robotic food prep and kitchen design optimized for robots rather than chefs. With these new technologies, as Kalanick says, “digitizing the physical world is my life’s work.”
As we reconfigure our physical spaces for robotic collaborators, expect new expectations and disruptions. Humans want physical infrastructure built for their comfort and efficiency, with an eye towards aesthetic beauty, or at least the familiar; robots function best with simplicity, easy transit and “beauty” expressed in utility not visual appearance.
The End of Software (As We Know It)?
The software ecosystem is already being re-shaped for the convenience of AI agents. AI agents, born from large language models (LLMs), are designed to pursue goals on behalf of users with some degrees of autonomy and reasoning capability. They also feature the ability to invoke and utilize computing resources such as browsers, websites, applications and data stores to help accomplish these goals.
Progress in this area has been rapid. The state of the art has quickly evolved from early demonstrations of agents navigating web pages to perform online shopping, a mode that visually approximates your grandparents learning how to use Amazon.com in 1999, tentatively clicking around, back-spacing and often invoking the wrong commands. Now, agents are capable of performing sophisticated workflows and traversing multiple applications by leveraging a critical artifact of modern computing, the application programming interface (API).
APIs have become a valuable and dominant way to connect applications to data sources and to each other. Think of them as on-ramps and off-ramps that connect highways to cities—and other highways. They were part of the arcane plumbing that lies beneath the foundation of our increasingly well-crafted software and workflows. Humans engaged user interfaces to command software and engage with outputs like dashboards, while APIs labored in the background.
Now, AI agents can command individual applications and autonomously compose those API interactions into multi-step workflows. In this world, APIs are not only the plumbing, they are the interface. Agents don’t require a beautiful canvas to function. Rather, they need to access applications seamlessly, go right to the heart of the data store or logic layer to perform an operation and then move on to the next step. Agents favor applications that are set up to help them navigate – modern APIs, machine-readable content that functions as a user guide to the platform, service-level guarantees, transaction capability and telemetry to assess effectiveness.
These shifting requirements signal the rise of “headless software,” platforms that are optimized for agents and that favor utilitarian interfaces over elegant design. This evolution has been well described by leading software entrepreneurs such as Dharmesh Shah, co-founder and CTO of HubSpot, and Aaron Levie, founder and CEO of Box. Levie has observed that “enterprises need to be able to ensure all of their software works across any set of agents they choose.”
With the rise of headless software platforms, SaaS applications act increasingly like data repositories that agents can traverse and stitch into complete workflows. To the extent this paradigm continues to emerge, it implies a kind of relegation of some classical software applications with their elegant interfaces designed to engage humans. So-called systems of record remain valuable as reliable, persistent stores of corporate data, but they become a watering hole along the agentic journey, not a destination. Consequently, the value proposition of classical (in other words, human-centric) software may shrink commensurate with this new role and its pricing power. Perhaps the entire pricing model shifts in favor of usage or outcome-based revenue models that better align software vendors with customers’ desire for value realization.
The Worldwide Agent Web
Websites are undergoing a similar transformation. Decades of work to perfect the human appeal of websites’ user interfaces and commerce sites allow us to browse and shop in a familiar manner, loading virtual shopping items into virtual carts, and seeking our own optimal combination of price, quality and availability. However, such designs may now be superseded by austere sites that allow agents to act efficiently on our behalf.
Each form optimizes for different functions and users. Humans browse the internet looking for images, drop-down menus, buttons, slider bars, folders, dashboards, and web forms that allow us to interact with computing in ways that are familiar and intuitive. But to AI agents, these features are distractions, barriers to their direct access to data, capabilities and logic they need to complete tasks. Humans want intuitive interfaces and appealing visuals. AI agents want “clean” API surfaces, a descriptive markdown file, and JSON schemas.
The transformation of the internet to more agentic interfaces is already underway. As Mathew Prince, the co-founder and CEO of Cloudflare, a leading internet infrastructure company, Cloudflare, said in June of this year, “Agentic traffic [is] growing so fast that bots have now passed human traffic online for the first time in the Internet’s history.”
As agents traverse the web on our behalf, the practice of Search Engine Optimization (SEO), through which websites compete for human visitation by tuning their appeal and seeking referral of users from Google and other discovery platforms, is giving way to Artificial Engine Optimization (AEO), which is designed to induce agents to promote, visit and even transact on websites by increasing their visibility and appeal to our digital delegates.
As with software, the features that optimize for agents differ significantly from the intricate features of modern internet sites built for human use. Commerce and content purveyors are scrambling to contribute their data to train LLMs and place themselves squarely in the transactional path of agents. The implications for online commerce are particularly significant: The parallel processing capabilities of AI can allow agents to comparison shop at scales and speeds far beyond the capacities of humans.
Millions of websites have advantages of legacy and incumbency. But over time, people’s loyalty to online brands may be supplanted by the efficacy of their interfaces for agents. Unlike humans, agents don’t shop habitually or “get used to” shopping on any given platform. Dynamic pricing may become more pervasive amid these accelerated shopping sprees. A logical outcome for this development could be the rise of auction pricing at scale, where agents put out “requests for proposals” for every sweater or light bulb purchase they make on our behalf and induce online stores to compete to win every piece of business in real time. Tokenized payments may rise in use for agent-based transactions, where immutability and speed are most desirable. By the same token, tolling infrastructure may emerge that permits agents to autonomously “pay for” access to content or other online resources as they pursue our goals. Both of these possibilities could meaningfully challenge the current economic structure of the web, where human-driven search lies at the heart of online monetization and existing payment methods and rails are necessary to consummate transactions.
How this Might Play Out
We are beginning to see evidence that our digital and physical spaces are evolving with the advances of AI and robotics. This raises a series of critical questions. How dominant will “machine spaces” become? What models for co-existence and control will emerge? Will “human spaces” become a rounding error too?
Many of these changes are nascent. As such, they will coexist with established architectures. In these early years, we should expect the development of “parallel universes.” Software and websites built for humans won’t vanish. Factories, warehouses and stores won’t suddenly close. But a new and different infrastructure will eventually emerge – one that is designed to serve our digital workers in contexts where humans may no longer be the primary actors.
We may begin to consider the ergonomics of agents and robots as much or more than we consider optimization for humans. While some may view this wistfully, it seems likely that these changes and the tremendous efficiency gains they promise may free up capital and creativity to build new, entirely human-centric architectures that are not burdened by the compromises necessary to accommodate people and machines but rather are tuned exclusively to our highest tastes and aspirations.
The irony is that a world redesigned for machines may eventually allow us to recover something more purely human. If factories, kitchens, warehouses, websites, and workflows become increasingly machine-native, then human spaces may be relieved of some of their utilitarian burden. We may build more places for beauty, reflection, play, learning, and community precisely because the machine world has absorbed more of the work. The danger is that human spaces become incidental. The opportunity is that they become sublime.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
Source link