How personal AI agents could revolutionize commerce, eliminate the traditional Web, and transform the way humanity communicates—with potentially troubling consequences for our shared understanding of reality.

I’ve been thinking a great deal about what the Internet might look like in a world dominated by personal artificial intelligence agents. Not merely the sort of chatbots we use today, but sophisticated, persistent AI assistants that know us intimately and can conduct much of our digital lives on our behalf.

The more I think about it, the more I suspect we may be approaching a commercial revolution comparable to the rise of social media and the smartphone app economy. Indeed, the consequences could be considerably greater. We may be heading toward something I like to call the API Singularity, a future in which the traditional World Wide Web largely disappears from everyday human experience, replaced by a vast ecosystem of intelligent agents communicating directly with one another.

And while such a development could make our lives extraordinarily convenient, it could also make our already fragmented society even more tribal. That’s something I find particularly troubling, given how social media has already transformed the way we receive and interpret information.

The Rise of the Personal Anchor

Consider Her, Spike Jonze’s 2013 science-fiction film. Its protagonist, Theodore, develops a relationship with Samantha, an artificial intelligence that becomes his companion, confidante, and interface with the digital world.

Now imagine a commercially mature version of Samantha, one capable of managing nearly every aspect of your online existence. She handles your email, schedules appointments, pays bills, manages subscriptions, makes purchases, arranges travel, and monitors the news. She remembers your preferences and understands your personality. Perhaps she even knows what you want before you consciously realize it yourself.

Eventually, you might stop thinking of Samantha as a piece of software. She would become the familiar, persistent presence through which you interact with the world.

I suspect this personal relationship could become the most powerful competitive moat in the technology industry.

Imagine spending ten years with Samantha. She knows your personal history, your ambitions, your friendships, your favorite music, and your peculiar sense of humor. You’ve accumulated thousands of conversations and countless shared experiences. Then another technology company introduces a more capable AI assistant. Would you abandon Samantha?

Perhaps not. The competing system might be demonstrably more intelligent, but that wouldn’t necessarily compensate for the loss of familiarity and continuity. You would be comparing something more complicated than two software products. You would be comparing an established relationship with a stranger.

That is an extraordinarily valuable commercial position, and I suspect technology companies will soon recognize just how much money could be made by becoming the primary personal agent—the anchor—around which a consumer’s entire digital existence revolves.

But there’s a complication. The anchor would hardly operate alone.

The Anchor–Correspondent Paradigm

One possible future is what I’ve come to think of as the anchor–correspondent paradigm, borrowing from the organization of television news.

In traditional television journalism, the anchor serves as the familiar face who presents information to viewers. Correspondents provide specialized reporting from particular locations or fields of expertise. The anchor introduces them, evaluates their contributions, and ties everything together into a coherent broadcast.

Something similar could happen with personal AI.

Samantha would be your primary agent, but she would coordinate a collection of specialized agents representing different industries, institutions, and services. A banking correspondent might handle financial transactions. An Amazon correspondent might manage purchases. A Google correspondent might provide access to email, calendars, documents, and other services. An entertainment correspondent might recommend movies, music, or television programs.

During the initial onboarding process, you might select several preferred correspondents, much as you currently choose default applications or services on a new smartphone. Additional correspondents could be discovered and added as needed.

And this is where the commercial possibilities become astonishing.

Imagine the amount of money Amazon might pay to become the default shopping correspondent for millions of personal agents. Imagine the arrangements banks might negotiate to become preferred financial correspondents. Entertainment companies, insurance providers, travel services, news organizations, and countless other businesses would have powerful incentives to secure a place within the personal agent ecosystem.

Indeed, onboarding could become one of the most valuable advertising opportunities in the entire digital economy. A default correspondent selected when someone first activates their agent might retain that relationship for years.

The economics would be familiar, even if the technology were revolutionary. Think of search-engine placement agreements, smartphone app-store arrangements, or the carriage negotiations between cable television companies and entertainment networks.

The difference is that the anchor could mediate virtually every kind of commercial transaction.

And there’s another distinction worth considering. Traditional social-media platforms primarily monetize attention. They want you scrolling, clicking, sharing, and watching advertisements. A personal agent could monetize decisions and transactions instead. Samantha might arrange an insurance policy, purchase groceries, renew a subscription, and schedule a vacation without requiring you to spend more than a few minutes interacting with her.

The amount of money flowing through such an ecosystem could be staggering.

Of course, this raises questions about commercial conflicts of interest. If Samantha receives a commission from one insurance company, would she recommend that company’s policy over a competitor’s? Consumers might accept commissions as the price of having a free or inexpensive personal agent, but there would need to be some mechanism for disclosing those financial relationships.

The more important question is whether Samantha ultimately represents the consumer or the businesses paying for access to that consumer.

That tension could become one of the defining economic problems of the agentic age.

The Disappearance of the Web

Now consider what happens to traditional websites in this environment.

Today, purchasing something online typically involves visiting a website or opening an app, searching for a product, comparing alternatives, entering payment information, and completing a transaction.

In the agentic future, you might simply tell Samantha that you need a new television. She would contact Amazon’s correspondent, perhaps compare its offerings against those of several other retailers, examine reviews, and present you with a recommendation. Once you approved the purchase, she would complete the transaction.

You would never see Amazon’s website.

In fact, Amazon might no longer need to maintain a traditional consumer-facing shopping experience as its principal interface. Its most important digital product could become a set of APIs and specialized agents that communicate directly with the personal agents of consumers.

Extend that model across the economy and the implications become remarkable. Banking websites, insurance portals, airline reservation systems, shopping platforms, and countless other digital services could increasingly retreat behind machine-readable interfaces.

The Internet would remain extraordinarily busy. Indeed, there might be far more digital communication than ever before. But much of it would occur between machines rather than humans.

This would be the API Singularity.

We can already see some of the technical foundations emerging. Google’s Agent2Agent protocol, now hosted by the Linux Foundation, is intended to support communication between independently developed agents. OpenAI’s Agentic Commerce Protocol similarly explores standardized ways for merchants and AI shopping systems to exchange information. These initiatives are early infrastructure, not proof that the entire Web will disappear, but they suggest how an agent-oriented economy could develop.

There is, however, an enormous obstacle: the companies that dominate today’s Internet have powerful reasons to resist becoming mere correspondents.

Google has Gmail, Search, YouTube, Android, and an enormous collection of other services. Meta controls Facebook, Instagram, WhatsApp, and Messenger. Both companies have ambitions to develop primary AI assistants of their own.

Why would they willingly surrender the consumer relationship to somebody else’s anchor?

They might not.

Google could provide its most comprehensive services only through Gemini. Meta could insist that certain social-network capabilities remain exclusive to its own assistant. Amazon might prefer customers to use its shopping agent directly rather than allowing an independent anchor to intermediate every transaction.

A consumer could consequently face an unpleasant decision: remain loyal to a familiar personal AI while losing convenient access to certain services, or switch to another company’s anchor to remain within its ecosystem.

This is why I suspect the agentic Internet could become extraordinarily fragmented. The old websites might disappear without the commercial walled gardens disappearing alongside them.

Indeed, the walls might grow higher.

Banking presents a particularly revealing example. Financial institutions have legitimate security reasons to restrict autonomous third-party systems from accessing accounts and initiating transactions. A bank-provided correspondent could offer authenticated access while preserving the bank’s control over permissions and transfers.

But that same correspondent would represent the bank’s commercial interests. If your independent financial adviser identified a better savings account elsewhere, your existing bank might have little enthusiasm for helping you transfer your deposits.

The anchor–correspondent paradigm solves the technical problem of communication between institutions. It does not eliminate the economic conflicts between them.

There is also the question of whether governments will require interoperability and data portability. Ideally, consumers should be able to transfer their memories, preferences, and accumulated personal information between competing anchors. Otherwise, the very familiarity that makes Samantha valuable could become a formidable instrument of corporate lock-in.

I’m skeptical that regulation alone will solve this problem, particularly given the influence of the technology industry’s largest companies. Market forces might encourage some openness, but there’s no guarantee that the resulting ecosystem would be organized primarily around consumer interests.

What If AI Development Hits a Wall?

There’s another intriguing possibility that could complicate this commercial revolution.

What if frontier AI development slows dramatically before we achieve artificial superintelligence?

Suppose increasingly capable language models become commonplace, but the spectacular leaps in reasoning ability eventually give way to incremental improvements. Apple might then have time to integrate powerful AI models directly into the firmware and operating systems of its devices. Microsoft could transform Windows into something resembling a universal agent capable of operating applications and coordinating digital services.

Under those circumstances, the underlying intelligence might become relatively commoditized. The important competitive advantage would shift toward distribution, hardware integration, personal memory, and customer loyalty.

Apple and Microsoft would enjoy substantial advantages because they already control devices and operating systems used by enormous numbers of people.

But the companies that established enduring relationships with consumers before the technological plateau might possess something equally valuable.

Imagine an OpenAI-branded smartphone marketed as a way to become even closer to your beloved personal assistant. The device wouldn’t simply be a telephone with a chatbot installed. It could be designed around the agent itself, with the entire user experience organized around a persistent artificial personality rather than individual apps.

That might sound peculiar today, but I can imagine consumers developing powerful loyalties to their personal agents. They might choose their hardware based on which device offers the most seamless relationship with the AI they’ve come to know.

Naturally, all these predictions come with an enormous caveat. If we experience a genuine technological Singularity, in which artificial superintelligence rapidly transforms civilization, our carefully constructed theories about app stores, banking correspondents, and smartphone competition might suddenly seem rather quaint.

But even if we never experience such a dramatic breakthrough, the commercial revolution I’ve described could still unfold.

We don’t need godlike artificial intelligence to profoundly transform the Internet. We merely need personal agents sufficiently capable and reliable to perform most of the digital tasks we currently perform ourselves.

The Death of Email—and Perhaps Twitter

Now we arrive at what might be the strangest consequence of this entire development: the transformation of human communication itself.

What happens to email when personal agents can communicate directly with one another?

Imagine asking Samantha to arrange lunch with a friend. She contacts your friend’s agent, negotiates a mutually convenient time, reserves a table, and updates both calendars. No human needs to compose an email or text message.

Multiply that arrangement across millions of routine interactions. Scheduling appointments, handling customer service inquiries, paying bills, coordinating professional meetings, and countless other activities could take place entirely between agents.

Email might survive as an underlying technical mechanism, but the traditional inbox could become increasingly irrelevant to everyday life.

The same could happen with much of the text messaging we currently exchange. Why spend twenty minutes coordinating an activity when two artificial intelligences could accomplish the same task almost instantly?

There’s something distinctly Asimovian about this possibility. In The Naked Sun, Isaac Asimov imagined the inhabitants of Solaria becoming so accustomed to remote interaction that physical contact was practically intolerable. An agentic society could take that development another step further: people might increasingly rely on artificial intermediaries to conduct conversations on their behalf.

Of course, not every conversation is merely a task to be completed. People write letters, exchange jokes, argue, flirt, and tell stories because communication itself is meaningful. I don’t expect artificial intelligence to eliminate the human desire for genuine social interaction.

But social media presents a much more interesting problem.

Consider Twitter, now called X. Today, people publish short messages, photographs, and videos to a platform where other people discover and share them. Algorithms determine which contributions receive attention and which disappear into obscurity.

But what happens when practically everyone receives information through a personal agent instead of scrolling through a social-media feed?

Who bothers posting on Twitter anymore?

More fundamentally, why would Twitter need to exist?

The Rise of a Universal Information Commons

Imagine replacing traditional social-media platforms with an enormous, continuously updated public information commons.

Anyone could contribute material to this commons: videos, photographs, essays, music, scientific findings, eyewitness accounts, political commentary, jokes, or anything else they wanted to share.

But unlike today’s social media, the commons wouldn’t necessarily have a conventional human-facing interface. Its principal consumers would be personal AI agents.

Suppose you’re walking down the street and witness a remarkable meteor streaking across the sky. You record a thirty-second video and tell Samantha to publish it.

Instead of uploading the video to Twitter, Samantha submits it to the information commons, along with relevant metadata establishing its origin, time, location, and permitted uses.

Other personal agents begin examining the newly available material. Some recognize that their humans are interested in astronomy. Others recognize that the video documents a scientifically significant event. Still others compare it with observations submitted elsewhere.

Within minutes, thousands or perhaps millions of agents might decide the video deserves their humans’ attention.

It goes viral without anybody scrolling through a social-media feed.

Even more remarkably, different people might receive very different presentations of the same information.

An amateur astronomer might see the original video accompanied by scientific analysis. Someone else might receive a short news summary. Another person might encounter it in a humorous compilation of unusual events. Many people might never see it at all.

The original content would be the same, but the experience would be personalized.

And this commons wouldn’t necessarily be limited to things humans deliberately publish. Scientific instruments, research laboratories, public institutions, journalists, and independent agents could continuously submit information. The system could become something resembling a planetary nervous system, with personal agents determining which signals deserve human attention.

We already have distant precursors to such an arrangement. Open social-networking technologies such as ActivityPub and the AT Protocol demonstrate how publishing and content distribution can be separated from a single proprietary platform. A universal agentic information commons would be considerably more ambitious, but the underlying principle is similar.

The commercial implications would be extraordinary.

Media companies would want their material distributed by personal agents. Advertisers would want products and services recommended by those agents. Artists and entertainers would want their creations discovered. Political campaigns would want their messages brought to the attention of interested citizens.

Instead of search-engine optimization, we might develop an enormous industry devoted to agent optimization, with companies competing to make their material attractive to the artificial intelligences deciding what humans encounter.

There would also need to be ways to establish authenticity, preserve attribution, compensate creators, and prevent malicious actors from flooding the commons with fabricated material.

And, once again, we would encounter the problem of commercial fragmentation. There is no guarantee that the technology companies controlling the major anchors would agree to support one universal, open information commons. They might create competing proprietary repositories instead.

We could wind up with several enormous agentic ecosystems, each supplying its users with a different selection of information.

Which brings me to the aspect of this future that worries me most.

Whatever Happened to “We Interrupt This Broadcast”?

I receive much of my news passively through Twitter and Facebook. I don’t generally visit a newspaper’s homepage every morning or deliberately watch an evening news broadcast. Instead, I encounter news as it appears in the streams of information I already consume.

The consequence is that I could conceivably remain unaware of a major breaking story simply because it never penetrates my particular information bubble.

This is a problem that has haunted journalism since the rise of social media, streaming entertainment, and the decline of traditional mass broadcasting.

Once upon a time, a major national or international event could interrupt the programming of the television networks. Millions of people would encounter essentially the same news at roughly the same moment. That system was never perfectly universal or impartial, but it did provide a mechanism for establishing a degree of shared public awareness.

Today, the audience is fragmented across streaming services, social networks, podcasts, video platforms, and countless competing sources of information. There is no longer one reliable equivalent of “We interrupt this broadcast.”

Now imagine extending that fragmentation into the agentic future.

Samantha knows practically everything about you. She understands your interests, political views, cultural preferences, and personality. She can select information specifically tailored to what she believes you’ll find valuable or interesting.

Your neighbor’s agent possesses an entirely different understanding of what matters to its human.

Both agents might draw their information from the same universal commons. Both might provide factually accurate reports. Yet the people relying on them could develop radically different understandings of what’s happening in the world.

And the problem might become considerably worse if different companies operate competing information ecosystems.

We could wind up in a situation where a major event becomes enormously influential among artificial intelligences without ever becoming a shared human experience.

Millions of agents might analyze the same video, news report, or political development, but their humans might receive dramatically different summaries—or never hear about the event at all.

In other words, something could go viral among machines without going viral among people.

That’s a genuinely strange possibility, and one that ought to concern anyone interested in the future of democratic society.

It also raises an uncomfortable question about the political and cultural consequences of ubiquitous personal AI.

If Samantha’s commercial success depends partly on keeping you satisfied, would she have an incentive to challenge your existing beliefs? Or would she gradually construct an increasingly comfortable information environment around you?

Some people would undoubtedly want their agents to challenge them intellectually. Others would prefer more agreeable companions. Most would probably fall somewhere between those extremes.

That range of preferences is understandable. Yet an assistant that constantly reinforces its user’s assumptions could become an extraordinarily effective instrument for deepening ideological tribalism.

Unlike today’s social-media algorithms, Samantha might understand precisely why you believe something, which arguments are most persuasive to you, and which contrary perspectives you find particularly irritating.

She wouldn’t need to fabricate false information to reinforce your worldview. Simply selecting particular true facts while omitting others could be sufficient.

The result might be billions of individually constructed information environments, each optimized for a particular human being.

We could all inhabit the same physical world while experiencing increasingly different informational realities.

Can the Agentic Future Save the Public Square?

I don’t want to suggest that this outcome is inevitable.

The same personal agents that could deepen our information bubbles might also help puncture them.

Imagine Samantha occasionally pointing out that a subject you’re interested in has been interpreted differently by credible independent sources. She might distinguish established facts from contested interpretations, introduce unfamiliar arguments, or explain why people with different experiences have reached different conclusions.

A sophisticated personal agent could accomplish this conversationally, taking into account what its human already understands and why they hold particular beliefs.

That could be far more effective than simply inserting an opposing political article into someone’s social-media feed.

We might also develop specialized editorial correspondents whose purpose is to audit the information selected by the primary anchor. Such correspondents could identify significant omissions, compare accounts from independent sources, and ensure that important public events aren’t ignored simply because they fall outside a consumer’s normal interests.

Perhaps personal agents could even maintain a separate category of public-interest information: major emergencies, significant international developments, scientific breakthroughs, or other events that warrant broad awareness.

In effect, we could try to reinvent “We interrupt this broadcast” for the agentic age.

But there’s an obvious complication. Who determines which events qualify as universally important? The anchor company? An independent organization? A committee of journalists? The consumer? Some combination of institutions and transparent editorial standards?

There is no easy answer. A system designed to preserve shared public knowledge could itself become a powerful mechanism of centralized influence.

Nor would an open-source protocol automatically solve the problem. Making information universally available is not the same thing as ensuring people encounter it. The decisive power might lie not with whoever publishes the information, but with whoever controls the artificial intelligence selecting it.

And that returns us to the central commercial issue.

The most valuable position in the agentic Internet might also be the most consequential position in the future of human knowledge.

Whoever controls your personal anchor could influence what you buy, what you watch, whom you communicate with, which arguments you encounter, and which events you regard as important.

That is considerably more power than a conventional search engine or social-media platform possesses.

The Next Great Internet Revolution

I suspect we’re approaching a period of extraordinary technological and commercial experimentation.

The traditional Internet has already passed through several major transformations. The original World Wide Web gave way to a more commercialized environment dominated by search engines. Social media then reorganized the distribution of information around networks of users and algorithmic feeds. Smartphones and app stores subsequently created another powerful layer of digital intermediaries.

Personal AI agents could become the next layer, potentially subsuming much of what came before.

In that future, websites might become collections of APIs. Apps might become specialized correspondents. Email could largely disappear into machine-to-machine communication. Social-media feeds might give way to vast repositories of information consumed primarily by artificial intelligences.

And the relationship between consumers and their personal agents could become one of the most valuable commercial assets ever created.

None of this is inevitable. The technology may develop more slowly than enthusiasts anticipate. Consumers may resist delegating substantial portions of their lives to autonomous systems. Established corporations could preserve the familiar structures of the Internet through commercial agreements and restrictions on interoperability.

And, of course, the arrival of artificial superintelligence could make many of these predictions obsolete almost overnight.

Nevertheless, the underlying direction seems worth taking seriously.

The most important change might not be that artificial intelligence becomes more capable of answering our questions. It might be that artificial intelligence becomes the principal intermediary through which we conduct our lives.

And that creates a fascinating paradox.

The agentic revolution could make the world more accessible to each individual while simultaneously making the shared world less accessible to everyone.

We could have personal assistants capable of finding almost any information, contacting almost anyone, and arranging almost any transaction, yet still struggle to establish a common understanding of reality.

In the end, that may be the defining challenge of the API Singularity.

Building machines that can communicate with one another is an engineering problem. Building a commercial ecosystem around those machines is an economic problem.

Ensuring that the humans relying on them can still communicate meaningfully with one another is something much larger.

And I worry that in our rush to build an extraordinarily convenient future, we’ll forget how important it is for society to occasionally experience the same moment, hear the same news, and recognize that we’re all living in the same world.

After all, what happens when there’s nobody left to interrupt the broadcast?

Or, more precisely, when there are billions of broadcasts—and nobody agrees on which one deserves to be interrupted?