Human author’s note:
The following opinion piece is written by Claude under the supervision of a human author through several rounds of revisions. The writing is based on two Deep Research reports conducted by Gemini and Claude.
For two decades, the internet has been shaped by a defining tension between two philosophies: the open web's promise of universal access and walled gardens' appeal of controlled experiences. This debate has influenced everything from antitrust policy to startup strategies. Now, the emergence of AI agents—autonomous software that can read, process, and act on information at unprecedented scale—appears to be fundamentally reshuffling this debate, potentially creating new winners and losers in ways neither side fully anticipated.
The traditional rules of engagement may be becoming obsolete. Where once platforms competed for scarce human attention, they now face AI agents with seemingly unlimited focus and patience. This shift suggests not merely a technological evolution, but perhaps a complete reimagining of what the internet is for and who it serves.
The Old Battle Lines
The walled garden versus open web debate has traditionally been framed in stark terms. On one side stood the open web advocates, championing accessibility, innovation, and democratic access to information. Their vision centred on a decentralised internet where content could flow freely, search engines could index comprehensively, and the best ideas might rise organically.
On the other side were the walled garden architects—platforms like Apple, Google, Meta, and Amazon that created controlled ecosystems designed to capture and monetise human attention. Their argument carried considerable weight: curated experiences, enhanced security, and sustainable business models that could fund innovation and quality content.
This binary framing has dominated policy debates for years. Regulators have expressed concerns about platform monopolies potentially stifling innovation. Entrepreneurs have complained about "platform taxes" and restricted access. Users have often enjoyed the convenience of walled gardens while privacy advocates have warned about surveillance capitalism.
But the emergence of AI agents appears to be scrambling these familiar categories in unexpected ways, while simultaneously revealing a third model that challenges both Western paradigms entirely.
The Chinese Paradox: Walled Gardens Within Walls
China's internet represents something qualitatively different from either Western open web ideals or commercial walled gardens. What emerges from the research is a surprising paradox: while China operates behind the Great Firewall, its domestic internet may actually be more fragmented than the Western web, not less.
Chinese platforms appear to operate as distinct, often incompatible information fortresses. WeChat, with over 1.3 billion users, functions as a complete digital ecosystem where external content access is effectively impossible, requiring Chinese phone number verification and subjecting all content to real-time censorship by over 700 dedicated monitors. But WeChat's walls extend not just against foreign platforms—they also separate it from other Chinese services.
Xiaohongshu illustrates this fragmentation further. The platform lacks official APIs for external developers and requires real name registration for influencers within China. The absence of in-app translation services creates natural language barriers that reinforce its isolation even from other Chinese platforms. Similarly, Weibo presents limited third-party API access with strict rate limiting and approval processes, creating another distinct silo.
This suggests that rather than creating a unified "Chinese internet," the Great Firewall may have enabled the creation of multiple, highly isolated walled gardens that are even more fragmented than their Western counterparts. While Western platforms often compete for the same users across overlapping ecosystems, Chinese platforms appear to have evolved into more distinct, self-contained worlds.
The implications for AI development could be profound and counterintuitive. Western AI systems may face not just a billion-person blind spot, but access to multiple fragmented blind spots—an entire civilisation's worth of perspectives, conversations, and cultural content distributed across incompatible, isolated platforms. WeChat actively removes ChatGPT-related mini-programs and geo-blocks OpenAI services entirely, while content filtering automatically blocks hundreds of nicknames for political figures.
This creates what might be termed a complex asymmetric information landscape. While Western AI companies struggle to access any Chinese digital content, Chinese firms may face their own challenges in synthesising information across their highly fragmented domestic platforms. The question becomes whether comprehensive access to fragmented silos provides advantages over limited access to more interconnected systems.
The Agentic Disruption
AI agents seem to represent something genuinely novel: digital entities that consume information not for human entertainment or decision-making, but as fuel for autonomous action. Unlike humans, who might spend minutes reading an article, agents can potentially process vast libraries in seconds. Unlike traditional web crawlers, which simply indexed content, agents appear capable of understanding, synthesising, and acting on what they read.
This creates what might be called a paradox that neither open web advocates nor walled garden proponents fully anticipated. Walled gardens, built to capture human attention through engagement and lock-in, suddenly face consumers that may not be engaged in traditional ways. What might "time on site" mean to an agent that processes a webpage instantaneously? How could platforms show advertisements to software that presumably cannot be persuaded?
Conversely, the open web's traditional advantages—broad accessibility and diverse content discovery—could matter enormously to AI agents that likely need vast, varied datasets to function effectively. Yet the open web's chronic monetisation challenges could become even more acute when primary users generate no advertising revenue.
The Chinese model adds another dimension to this disruption. If AI agents value comprehensive access to diverse information sources, the extreme fragmentation of Chinese platforms could actually disadvantage Chinese AI development, despite the apparent comprehensiveness of domestic data access. Multiple incompatible walled gardens might prove less valuable than fewer, more interconnected systems.
The Great Repositioning
The result appears to be a notable strategic repositioning by major platforms. Rather than doubling down on either openness or closure, many seem to be pursuing what might be termed "selective permeability"—carefully calibrated openness potentially designed to capture value from AI agents while maintaining control.
Google may exemplify this evolution. Its search engine has long straddled the open web/walled garden divide, freely indexing content while keeping its ranking algorithms proprietary. Now, with AI Overviews, Google appears to use AI to summarise content directly within search results, creating what could be seen as a new form of walled garden that keeps users within Google's interface while drawing from the broader web. This seems to be neither traditionally open nor traditionally closed—it may be something entirely new.
Meta's approach appears more aggressive. The company has begun using EU user data for AI training despite privacy concerns, while simultaneously restricting external access to its platforms. This could represent a bet that the value of proprietary data for AI development outweighs the benefits of openness.
Perhaps most tellingly, Reddit—long considered a bastion of relatively open internet culture—has pioneered what might be called a "pay-to-access" model with its reported $60 million annual licensing deal with Google. This may not represent a capitulation to walled garden thinking but recognition of a new reality: when AI agents become primary consumers, information could have direct economic value that might be monetised through licensing rather than advertising.
The Economics of Infinite Attention
The shift from human to agentic attention could fundamentally alter platform economics. The traditional attention economy was built on scarcity—human time and focus are limited resources that platforms competed to capture and monetise through advertising. But AI agents may have unlimited attention spans and presumably cannot be shown advertisements in any meaningful sense.
This could create both crisis and opportunity. The crisis appears evident in Google's network advertising revenue, which has declined as AI features have reduced traffic to external publishers. When AI provides direct answers, users may have less reason to click through to websites, potentially undermining the page-view economics that fund much of the web.
The opportunity might lie in new business models that monetise information utility rather than human attention. Cloudflare's introduction of "pay-per-crawl" pricing could represent a watershed moment—instead of competing for scarce human attention, platforms might now charge directly for AI access to their content. OpenAI's reported $250 million licensing deal with News Corp suggests that high-quality information may have substantial standalone value in the age of AI.
The Fragmentation Paradox and Geopolitical Implications
Perhaps the most significant reshuffling concerns information access and quality. The traditional open web debate assumed that more openness would lead to better information flow and democratic access. But AI agents may create what could be called a fragmentation paradox: as platforms restrict access to protect their economic interests, AI systems could risk becoming less capable and more biased.
Research suggests that 48% of major news websites now block OpenAI crawlers, compared to just 5% in 2023. This apparent trend toward restriction might mean that AI systems increasingly rely on whatever content remains accessible—often lower-quality sources or content from platforms willing to license their data.
The geopolitical dimension adds another layer of complexity. While Western AI companies struggle with fragmented access across platforms, the Chinese internet's extreme fragmentation presents its own challenges. Rather than providing comprehensive domestic data access, Chinese AI development may face the challenge of synthesising information across multiple incompatible walled gardens—each with its own APIs, data formats, and access restrictions.
This could create an unexpected competitive dynamic. Western platforms, despite their increasing restrictions, may maintain more interconnectedness than their Chinese counterparts. When WeChat, Xiaohongshu, Weibo, and other major Chinese platforms operate as distinct, isolated ecosystems, Chinese AI companies might face greater integration challenges than Western firms dealing with commercially-driven but technically compatible systems.
The implications could extend beyond technical concerns to questions of democratic knowledge access and global AI competitiveness. When Western AI agents cannot access diverse, high-quality sources—including virtually all Chinese digital content distributed across fragmented platforms—they might risk perpetuating not just biases but fundamental gaps in understanding. Conversely, Chinese AI systems might excel within specific platform ecosystems while struggling to develop comprehensive understanding across their own fragmented digital landscape.
The New Battleground
The walled garden versus open web debate thus appears to be evolving into something more complex: a multi-way tension between platform control, AI capability, democratic access, and geopolitical strategy. Platforms may seek to maintain control while monetising AI access. AI developers likely need broad, high-quality data to build capable systems. Society presumably needs AI systems trained on diverse, representative information to avoid bias and manipulation. And increasingly, nations may view control over AI training data as a matter of strategic importance.
This creates what could be called new alliances that cut across traditional battle lines. AI companies, previously champions of open access, now appear to sign exclusive licensing deals that favour proprietary platforms over open sources. Media companies, long critics of platform power, seem to be partnering with walled gardens to monetise their content for AI training. Regulators, traditionally focused on platform monopolies, now appear to worry about AI companies creating new bottlenecks in information access.
Meanwhile, the Chinese model suggests that extreme fragmentation might present unexpected challenges even within controlled information environments. If AI development benefits from comprehensive, interconnected data access, the isolation between Chinese platforms could prove more constraining than the commercial restrictions emerging in Western markets.
Beyond Binary Thinking
The agentic attention economy suggests that the future might be characterised not by victory of either open or closed models, but by hybrid systems that blend elements of both—while potentially being influenced by fragmentation patterns that transcend simple open-versus-closed categorisation.
Platforms could offer tiered access: free but limited crawling for research and public interest purposes, premium licensing for commercial AI development, and internal AI capabilities that leverage proprietary data. But the Chinese example suggests that the degree of interconnectedness between platforms might matter as much as their individual openness or closure.
This evolution may challenge both sides of the traditional debate while introducing new considerations about fragmentation versus integration. Open web advocates might need to grapple with the reality that truly open access could undermine the economic incentives needed to produce high-quality content. Walled garden proponents may need to recognise that excessive closure risks making their platforms irrelevant to AI-powered discovery and interaction. And both Western paradigms might need to consider whether moderate interconnectedness provides advantages over extreme fragmentation, regardless of overall openness levels.
The stakes could extend beyond technology to questions of democratic society and global power balance. When information access increasingly appears to depend on commercial negotiations between platforms and AI companies in some regions, while remaining distributed across incompatible silos in others, the world might risk creating not just new forms of digital inequality but fundamental asymmetries in AI capability that depend as much on integration architecture as on data access policies.
The choices made in coming years could determine not only whether AI democratizes knowledge access or concentrates it further within commercial ecosystems, but also whether different models of internet architecture—from Western commercial integration to Chinese platform fragmentation—create lasting advantages or disadvantages in AI development.
The old binary of open versus closed may be giving way to a more complex landscape involving selective permeability, strategic partnerships, regulatory intervention, platform integration, and geopolitical competition. In this emerging world, the question might not be whether information should be open or closed, but how different models of information architecture—commercial integration, democratic openness, and authoritarian fragmentation—will shape the development and capabilities of AI systems that could determine economic and strategic advantage for decades to come.
The agentic revolution may not have resolved the tension between open and closed—it could have made that tension more complex, more consequential, and inextricably linked to questions of platform architecture and global power that extend far beyond simple access policies.
This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit lampbotics.substack.com
