Muse, Dots, Duet, Spark, Copilot, Grok Bot. What was once the domain of tech specialists in San Francisco, New York, and London has now entered the mainstream. Expert users have deployed agentic AI—both individually and in coordinated swarms—to handle everything from building custom applications and managing businesses to booking services and navigating healthcare decisions. That exclusive window is closing fast.
Spreading access to AI agents promises genuine benefits: freed-up time, reduced labor burdens, and democratized capabilities once reserved for the tech-savvy. Yet beneath this appealing surface lies a more troubling prospect. A handful of corporations could seize control of agentic AI technology, creating a monopoly or oligopoly that concentrates extraordinary power in private hands. A recent report, "New Gateway, Same Gatekeepers," from the Open Markets Institute Europe examines this threat, maps out the economic and democratic dangers, and proposes solutions.
Personal agents: the endgame for consumer AI
While AI researchers continue debating artificial general intelligence, the industry has moved on to building products it can sell. Consumer engagement with AI had remained largely confined to chatbot conversations until recently. That shifted starting in mid-2026, when OpenAI, Anthropic, Meta, Google, Microsoft, and xAI all rolled out accessible personal AI agents capable of taking action on behalf of users.
The corporate vision is straightforward: develop and monetize a personal assistant that automates as many routine activities as possible. The goal is to position themselves as a universal digital intermediary—a single platform handling groceries, clothing purchases, holiday bookings, government interactions, home control, news consumption, and communication with friends and family. Rather than navigating dozens of apps and websites, users would rely on one interface, accessible via text or voice, for nearly all digital tasks.
The appeal is undeniable. People could reclaim time for relationships, personal interests, and meaningful work. Yet this vision simultaneously raises profound concerns about individual autonomy, economic competition, and democratic integrity.
New digital gatekeepers with unprecedented power
The risk is real: one or a small cluster of companies could come to dominate agentic AI, establishing a monopoly or oligopoly. In that scenario, these firms would become the gatekeepers determining what people purchase and from whom, which news reaches them, and how they connect with their social circles.
While fair stewardship is theoretically possible, commercial incentives and the absence of competitive pressure to treat users well would likely push these corporations toward exploitation. Armed with vast personal data stores and acting as a digital intermediary for users' actions, they would enable surveillance and exploitation—both commercial and political—at unprecedented scales.
A concentrated market would generate gatekeeper dynamics far exceeding those of Amazon, Google, and Apple today. An agentic AI gatekeeper could impose steep commissions on merchants selling through its agents. It might prioritize recommendations favoring paying partners and advertisers, transforming the entire economy into a pay-to-play system. Surveillance-based pricing at extreme levels would become feasible as monopolists channel unprecedented volumes of personal information through their agents.
Beyond commercial harm, concentrated agentic AI control would concentrate political power in private hands. People increasingly turn to AI agents and assistants for news, information, and political analysis, then base decisions on that input. Whoever operates these tools shapes what information users encounter and how it is framed, creating avenues for both direct and subtle influence.
In an AI oligopoly, that authority rests with a small number of private firms or state-influenced actors. Agentic AI extends political power beyond information control into potential coercion. Companies controlling agents could wield enormous influence over significant economic sectors and access to vital services—power they could deploy for political ends.
The road to concentrated agentic AI
A handful of private actors could gain unparalleled influence over vast populations through agentic AI, influence they are likely to exploit. In a genuinely competitive landscape where switching between agents is frictionless, these dangers diminish: competitive forces discourage abuse because users can defect to alternatives. But if a few firms dominate and users face switching costs, the hazards escalate dramatically.
Personal agents are not entering a competitive vacuum. They arrive in a world where tech giants already command sprawling digital empires spanning hardware, operating systems, browsers, search engines, email, messaging, social platforms, productivity tools, and online retail. These incumbents bring massive capital, distribution networks, and integration capabilities that newcomers cannot replicate.
While Apple continues refining Siri, Google, Meta, and Microsoft have each launched personal agents that leverage their existing market dominance across multiple sectors.
How five tech giants tie their AI to the rest of their products
Each company's AI assistant draws on products it already owns, and many of those products steer users toward the company's own services.
Google — AI assistant: Gemini "Personal Intelligence" — 29 products across hardware, browsers and operating systems, productivity (Docs, Sheets, Slides, Drive, Keep, Calendar, Tasks, Translate), communications, search and discovery, marketplaces and e-commerce, and media and entertainment.
Microsoft — AI assistant: Copilot — 18 products across hardware, browsers and operating systems, productivity (Word, Excel, PowerPoint, OneNote, Outlook, Calendar, OneDrive), communications, search and discovery, software development, and social and professional networking.
Meta — AI assistant: Muse — Products across hardware, social networking (Facebook, Instagram), communications (WhatsApp), and marketplaces and e-commerce (Facebook Marketplace, Facebook Shops, Instagram Shopping, Meta Pay).
Apple — AI assistant: Siri / Apple Intelligence — 29 products across hardware (MacBook laptops, Desktop PCs, iPhone, iPad, Apple Watch, AirPods, HomePod, Apple TV), browsers and operating systems, productivity (Pages, Numbers, Keynote, Notes, Mail, Calendar, iCloud Drive), entertainment, communications, commerce, software development, and social networking.
Source: Open Markets Institute Europe, "Same Gatekeepers, New Gateway: AI Agents, Monopoly Power, and Democratic Risk," Nov. 9, 2026.
Distribution provides the first advantage. Existing platforms allow these firms to instantly reach millions or billions of users. Meta's Muse is embedded directly into WhatsApp, which has 3 billion monthly users, alongside its standalone app. Google's Spark agent currently operates through Gemini apps, but Google has distributed AI services via Search, Chrome, Android, and Workspace, and will almost certainly do the same for agents. Microsoft is promoting Copilot through Microsoft 365 apps, GitHub, and Windows.
Integration amplifies this edge. Meta's Muse enjoys privileged access to Facebook, Instagram, and Facebook Marketplace in addition to WhatsApp. Google's Spark integrates deeply with Workspace and spans Gmail, Maps, Flights, Shopping, and YouTube. Microsoft's Copilot connects to Outlook, Teams, and the Office suite. Competitors cannot match this integration.
Beyond leveraging existing monopoly power, the technology itself may drive concentration. Two mechanisms stand out.
First, network effects: if agents lack interoperability, the more developers and service providers build for a particular agent, the more valuable it becomes to users, attracting additional users and making that agent the platform developers prioritize. Competitors face a thinner ecosystem and weaker incentives for users to switch.
Second, data feedback loops: as an agent learns from interactions, it improves with use. The provider with the largest user base accumulates the most data, builds the superior agent, and attracts even more users.
Specific design choices can also lock users into a single provider's agent, particularly those involving personalization and memory. An agent familiar with your preferences, work habits, contacts, and past choices is far more useful than a blank slate.
Today, that context remains under the provider's control rather than the user's. If companies do not permit context export in a portable format, switching agents forces users to start over. The longer someone uses an agent, the costlier departure becomes.
These factors together could push the market toward concentration. They also suggest that newer entrants like OpenAI and Anthropic retain potential for dominance, despite lacking the distribution and ecosystem advantages of established tech giants.
New risks, known solutions
The core question is straightforward: whose interests do these agents serve? As markets concentrate, the answer increasingly diverges from users and the public good.
Policymakers can address this through two approaches.
First, steer markets away from monopoly by minimizing lock-in and maximizing interoperability and data portability. Technology markets have proven solutions: open standards and protocols. The Model Context Protocol (MCP), which enables agents to connect to third-party services, represents a promising step. Regulators can build on this by developing and mandating technical standards that encourage pro-competitive innovation. The UK's Open Banking Standard demonstrates this approach's effectiveness.
Simultaneously, Big Tech companies wielding monopoly power in adjacent markets must not exploit that dominance to gain unfair advantages in agentic AI. The EU moved decisively in Meta's case, requiring it to allow competitors to distribute AI assistants via WhatsApp. It has taken similar action regarding Google's integration of AI into Android.
The UK's Competition and Markets Authority has flagged Microsoft's Copilot integration into its products as problematic and is examining interventions mandating interoperability, data portability, and fair choice architecture.
These represent only three instances of what appears to be a systemic industry pattern. Competition regulators must confront this epidemic of self-preferencing proactively, not reactively. Policymakers have made the "wait and see" error repeatedly in tech. Agentic AI cannot afford another repetition.
Competition enforcement alone cannot resolve the inherent conflicts of interest in agentic AI. Given the sensitivity of agent functions and their breadth of application, regulation must impose transparency and accountability on providers. Users, supported by regulation, must be able to verify that agents serve their interests. Legal obligations comparable to those applied to human agents for centuries must be adapted for AI. This encompasses disclosure of conflicts of interest, binding duties of care, and prohibitions on using sensitive personal data for commercial purposes, all binding on companies designing and marketing these agents.
Agentic AI's future will be shaped by policy decisions taken today
The promise of agentic AI is tangible. It could help people spend less time on routine tasks and more time on what matters. More significantly, agentic AI could redistribute power in the digital economy: helping publishers capture more value from their content, helping consumers understand and negotiate service terms, and helping people customize their online experiences by canceling subscriptions, managing feeds, and countering dark patterns and engagement-maximizing tactics.
If regulators succeed, the transformation could be substantial. People would have a capable assistant working for them rather than for a corporation: one that operates across services regardless of its creator and remains accountable to user interests. Small businesses and independent creators could reach customers and earn revenue without passing through a handful of gatekeepers. Time reclaimed from digital chores could go toward work, family, friends, learning, and hobbies. That future is within reach, but only if policymakers build it intentionally.



