The swift evolution of artificial intelligence is transforming the web as we know it—and Microsoft CTO Kevin Scott is at the center of this profound shift. As Scott explained in a recent conversation with The Verge, this is more than just another tech upgrade. It’s a reimagining of the architecture underlying digital discovery, interaction, and creativity. The stakes couldn’t be higher: can AI save the web from the economic and creative collapse that threatens so many online creators and businesses, or will it accelerate that decline by centralizing power in the hands of a few massive players?
The Web’s AI Tipping Point
For decades, the web has functioned under a clear set of incentives. Websites provided free content discoverable via centralized search engines—mostly Google and Microsoft’s Bing. In return, they earned revenue through ads, subscriptions, or other monetization models linked directly to web traffic. But recent advances in AI, especially in natural language processing and agentic software, are turning this model on its head.The issue is painfully clear: as AI-powered answers and chatbots proliferate, users increasingly get information directly from AI platforms rather than visiting original websites. Content creators lose out on valuable traffic and the ability to monetize their expertise. Google’s own search innovations are reducing click-through rates for countless sites, while alternative discovery platforms like TikTok or YouTube siphon more attention away from the open web. If left unchecked, this trend could hollow out the diversity and financial sustainability of the internet.
Enter Agentic AI and the Promise of MCP
Scott’s vision—articulated both in his role as CTO and as a participant in the industry dialogue—is that AI doesn’t have to destroy the web. Instead, it can strengthen it by decentralizing the very fabric of search and information retrieval.At the center of this vision is the Model Context Protocol (MCP), an open standard initially developed by Anthropic but now supported by a growing ecosystem including Microsoft, OpenAI, and even Google. Unlike current search paradigms that rely on a handful of massive web indexes, MCP allows any website to make specific content or services available to AI agents in a controlled, transparent, and programmable way.
Instead of AI bots awkwardly “clicking around”—attempting to mimic human browsing in a brittle and often blocked manner—MCP enables a structured, open communication channel. Site owners can decide exactly what their endpoint exposes, under which terms, and with what monetization schemes. That means it isn’t just a technical protocol; it’s a foundation for rebuilding the web’s broken business model.
The Natural Language Web and Local AI Search
To spur adoption, Microsoft has announced new tools—demonstrated at its Build conference—that make it trivial for websites to integrate AI-powered natural language search. Powered by MCP, even smaller sites can quickly add advanced search and agent compatibility, running on open-source or commercial models alike.The results, Scott argues, are striking. Visitors can now find exactly what they’re looking for in plain English, even on small or poorly indexed sites, reducing the reliance on Google for discovery. Tripadvisor reportedly went from discovery to internal demo in less than a week, showing just how accessible this technology is becoming.
By flipping the traditional model—enabling search to occur locally, under each site’s control, and federated at the agent level—this approach would dramatically lower the entry barrier for smaller players and allow for a renaissance of digital competition.
Economic Realities: Can Creators Survive?
Yet Scott is candid about the complexity of the economic incentives. With AI intermediating more interactions, how will content creators and site owners get paid? So far, the collapse in referral traffic has not been accompanied by a new monetization model.MCP is intentionally “opinionatedly neutral”—it specifies how agents talk to sites, not how anyone gets paid. This agnosticism is both a strength and a risk. On the upside, it gives each creator the chance to experiment: perhaps by requiring user authentication, paywalls, subscription checks, or creative ad models. New forms of microtransactions or agent-mediated commerce could emerge, echoing the early days of the internet when open protocols like HTTP allowed for permissionless innovation.
On the downside, the lack of a de facto economic standard leaves everyone improvising. As AI-driven summarization and answer bots get better, the traditional logic of “create content, attract eyeballs, monetize traffic” grows shakier. Scott admits: “I would love to have a way not to spend so much of my time worried about traffic referrals, period. I would love to spend more of my time building up an authentic relationship with people who might be interested in my products and services.”
For many creators, social media and community-building now feel more like running a local store with loyal customers—direct relationships matter more, and algorithmic gatekeepers less. But not every business can pivot so easily. E-commerce remains a major use case for independent sites, as it allows them to escape “platform taxes” and retain control of their transactions. However, for pure publishers and artists, the way forward is less certain.
The Agentic Web: From Synchronous to Asynchronous Discovery
One of the most powerful ideas Scott advances is the shift from synchronous, attention-demanding web browsing to asynchronous, agent-driven research. Today, users must actively browse, compare, and transact—each site fighting for seconds of attention. In the agentic future, people could delegate research (“find me the best 10 pottery kilns for under $1000”), product comparisons, or even entire purchasing decisions to persistent digital agents.This isn’t science fiction. Early agentic experiences already exist in software development, where code-generation and automation agents have achieved genuine product-market fit for programmers. For other domains—shopping, travel, business services—the challenge isn’t vision, but scalable execution. As standards like MCP and projects like Microsoft’s NLWeb mature, the groundwork for this new mode of digital interaction is being laid.
Importantly, agents provide broader “reach” for content: they can proactively match user needs to offerings, even while the user’s attention is elsewhere. This expands the universe of possible interactions—if (and it’s a big if) creators and agents’ ecosystems can agree on fair, transparent terms for access and compensation.
The Battle for Standards: Openness, Ubiquity, and Power
The debate around MCP and agentic web protocols harks back to earlier battles in web history. Just as HTTP, RSS, or robots.txt enabled new forms of distribution and control, the current moment is charged with questions about centralization, lock-in, and platform power.Scott draws conscious parallels between MCP/NLWeb and the early days of the internet, insisting that the simplicity and openness of these protocols are central to their potential success. Unlike proprietary standards—of which Microsoft itself was once a notorious proponent—the minimum friction and permissionless innovation promised by MCP could allow small developers and businesses to regain agency. “Sometimes there is a real problem in an ecosystem where the simplest solution that everybody can choose to adopt is the winner because we all win because ubiquity is the thing that really matters,” Scott argues.
But will the largest players—Google, Meta, Apple—embrace these open protocols, or seek to wall off their own vertical ecosystems? Apple’s App Intents parallels MCP within the iOS world, but is tightly controlled. Closed platforms, from TikTok to Facebook, may have no incentive to make themselves agent-visible and interoperable. Much rests on whether users begin to demand agentic experiences and reward openness in the market—a dynamic that’s still playing out.
Search’s Changing Power Structure
The web’s traffic—and, by extension, much of its economic and cultural power—has long been organized around Google. Scott observes that the rules are changing: as Google’s search model morphs and direct answers proliferate, many web creators are seeing less referral traffic than ever before. Organic search is no longer a reliable engine of discovery. Instead, social media, influencer channels, and specialized communities are becoming the main gateways to audience attention.Scott frankly admits he gets little meaningful traffic from search to his own small e-commerce venture. For many, the old “SEO game” has become increasingly abstract and less rewarding. In this context, agentic systems offer a fundamentally different kind of opportunity—the promise to be “agent visible” and to participate directly in new, value-generating transactions.
Still, this transition is fraught with uncertainty. Will agent-based discovery and commerce pay off for niche creators, writers, and small businesses? Or will the web’s fragmentation deepen, with only well-backed organizations able to stand out in the agentic era?
Risks and Roadblocks: Commercial, Social, and Creative
While the optimism is infectious, it’s hard not to see the landmines ahead. Some of the most pertinent risks include:- Erosion of Creator Incentives: If AI-driven summarization further reduces website visits, many sites—especially those relying on ad impressions or loose affiliation—may disappear entirely. Without robust mechanisms for attribution, payment, and two-way interaction, creators could be left uncompensated for their work, exacerbating a “tragedy of the commons” for digital culture.
- Fragmentation of Discovery: As social media, closed apps, and custom agentic experiences gain prominence, the open web could become increasingly isolated. If leading platforms refuse to participate in MCP-like standards, AI agents will have less information to work with, reducing their usefulness and reinforcing silos.
- Technical Immaturity: Despite promising demos and early adoption, agentic AI systems currently show real traction mostly in domains like software development. In most consumer and business contexts, robust, scalable use cases for agentic AI are only beginning to emerge. As Scott readily admits: “Outside of software engineering and outside of demos, I haven’t concretely seen the thing that works.”
- Persistent Monopoly Power: Even with open protocols, the biggest cloud providers and AI vendors still wield enormous influence. Ubiquity and interoperability are not guaranteed just because a protocol is open; history shows that business leverage and default choices matter greatly. The risk of new forms of centralization, inequity, or “embrace and extend” strategies cannot be dismissed.
AI’s Creative Dilemma: Ownership, Compensation, and Agency
No discussion of AI and the web is complete without addressing the creative economy—and the lawsuits, anxiety, and debate over generative AI’s use of original content for training and output.Scott, himself an author (with a presciently topical book, “Reprogramming the American Dream”), takes a nuanced stance. While he personally feels comfortable with his own work being freely summarized, he acknowledges: “If you spend all of your time and dump your heart and soul into making a thing, you deserve to be compensated for it. There’s a lot of different ways to be compensated for a thing…”
The central tension, as Scott describes, is between the transformative power of AI—especially in fields with high social utility like healthcare—and the legitimate claims of creative professionals for fair treatment and compensation. The U.S. Copyright Office, as Scott notes, is wrestling with this distinction: there is likely societal value in some types of use (e.g., medical model training), but competitive, commercial substitution for expressive creative work is crossing a line.
Scott believes the move toward open, agentic protocols may actually enable more flexible and creator-driven business models. For example, just as a subscription might be required for human users, agent access via MCP could be tied to payment, DRM, or other conditions. Technical advances in tracking, focus on attribution, and selective permissions could all help.
Yet, the industry has yet to deliver widespread, scalable models for compensating creators in the AI era. Litigation and controversy linger, and many artists and publishers are deeply skeptical of AI’s promises.
Technical Trends and the Burden of Hardware
The hardware side of things, often invisible to consumers, weighs heavily on Scott’s mind. Progress in AI has been propelled not only by better algorithms and more data, but by explosive hardware capacity—especially graphical processing units (GPUs). While new hardware generations continue to deliver “2X improvement in price performance,” Scott emphasizes that the real scale jumps are coming from software optimization. Techniques like data type compression, prompt routing, and caching allow existing systems to do far more with the same or even less raw compute power.This is crucial, since GPU supply remains a bottleneck for both research and commercial deployment. As workloads grow and more companies pursue advanced AI, the allocation and management of computational resources has become a central strategic issue for Microsoft and the cloud industry at large.
Notably, Scott is skeptical of the hype around “artificial general intelligence” (AGI), noting the lack of a precise definition and the myriad technical, social, and policy barriers that remain. For now, the focus is on tangible, near-term improvements, not science fiction breakthroughs.
The Open Web’s Next Act
So where do we go from here?Scott is clear-eyed but hopeful: the “deal of the old web” may not be totally over, but if it’s to survive in a form that still fosters creativity, diversity, and economic opportunity, serious architectural changes are needed. Open protocols like MCP and NLWeb represent an attempt to reimagine how content, services, and business models interact in a digital ecosystem increasingly shaped by intelligent agents and AI intermediaries.
Whether this will be enough—whether the incentives will be broad and compelling enough to reverse or stabilize the loss of web diversity—is impossible to predict. Much will depend on both the pace of technical iteration and the willingness of market participants to prioritize openness, fairness, and collaboration over short-term walled gardens and centralization.
Critical Analysis: Strengths and Risks
Strengths
- Openness and Flexibility: By designing protocols that let creators decide what to expose and under which terms, MCP avoids the pitfalls of one-size-fits-all solutions. Its simplicity encourages adoption.
- Decentralization: Local AI search and agentic interaction reduce dependence on centralized search indexes, potentially democratizing discovery and opportunity.
- Alignment with Market Trends: As users shift away from traditional search and toward social/AI-driven recommendations, agent-based interoperability could restore some discoverability to smaller or independent sites.
- Foundation for Innovation: Permissionless, open protocols enable rapid experimentation and the emergence of entirely new products and business models.
Risks
- Lack of Immediate Monetization: The protocol’s neutrality leaves creators adrift in searching for new economic models. Without reliable referral or direct compensation, the incentive to build or maintain quality content remains unclear.
- Platform Lockout: Major platforms (social, commerce, apps) may choose to wall themselves off, reducing the comprehensiveness and effectiveness of agentic AI systems.
- Fragmentation and Complexity: Multiple competing standards, coupled with technical immaturity, could confuse both creators and consumers, delaying adoption.
- Monopoly Dynamics: Large cloud and AI vendors may end up controlling the ecosystem anyway, even if standards are nominally open.
- Creative Concerns Remain: Although technical means exist for controlling access and monetization, widespread creator skepticism and legal ambiguity about AI’s use of copyrighted or original content still impede trust and participation.
Conclusion: A Pragmatic Path Forward
Scott’s argument—that AI can be the web’s savior rather than its nemesis—carries weight but warrants caution. The technical opportunities are real; the business and creative risks equally so. Crucially, the move to open protocols like MCP must be matched by serious, sustained work on fair attribution, compensation, and governance. Technical fixes alone will not resolve the tension between innovation and economic security for creators.The agentic web is not inevitable, but its core principles—openness, interoperability, permissionless innovation—harken back to the best days of the early internet. Getting there will be neither quick nor easy. But as Scott emphasizes, it is in these rare moments of architectural transformation—when “a bunch of hard things become easy” and the doors of participation swing wide—that entire digital economies can be reborn. Whether AI will ultimately save the web or contribute to its contraction rests not on technology alone, but on how wisely—and how fairly—we design the incentives, protocols, and relationships at its heart.
Source: The Verge Microsoft CTO Kevin Scott on the birth of the agentic web
