For seventeen years, every app store has agreed on one number: the platform takes thirty percent. Apple set it in 2008, Google copied it, and an entire industry of developers learned to resent it. At Build 2026, Microsoft walked on stage and broke that consensus in a single sentence, announcing a Windows Agent Store that pays developers an 85 percent revenue share. The pitch is not subtle. Microsoft is trying to buy the loyalty of every AI agent builder on the planet before Apple, Google, or OpenAI can lock them into a 70-30 split, and it is using the one asset none of them can match: 400 million Windows and Microsoft 365 seats that agents can now live inside natively.
What Actually Happened
Microsoft unveiled the Windows Agent Store alongside the Windows Agent Framework, a pairing meant to make AI agents first-class citizens of the operating system rather than browser tabs bolted on top. The store's defining term is the revenue split: developers keep 85 percent of what they earn, against the 30 percent cut Apple and Google have charged for years. Distribution rides on a new commercial bundle, an M365 E7 tier paired with an Agent 365 SKU priced at 99 dollars per user per month, and Microsoft's own modeling suggests that at just 10 percent adoption across its 400 million M365 users the program would throw off roughly 4.8 billion dollars in annual revenue.
The technical foundation is the Windows Agent Framework, released under an MIT license. It ships an Agent Registration Service, a local daemon that keeps agents alive, monitors their health, and handles versioning; a Cross-Agent Communication Bus built on gRPC so agents can signal and coordinate with one another; and a Memory Service that persists context and user preferences across sessions. Once registered, an agent can surface anywhere the user already works, appearing in the Windows taskbar, inside Copilot, and within Teams and Outlook. Microsoft is not asking developers to build a destination app. It is letting them inject capability directly into the surfaces a billion people open every morning.
Launch partners arrived to make the abstraction concrete. Adobe demonstrated an InDesign agent that automates layout and design workflows, and Zoom showed a meeting agent that summarizes calls and spins up follow-up tasks without a human opening a single menu. Governance came baked in through a layer Microsoft calls AgentGuard, which provides role-based access controls, data-loss prevention, and audit logging so that IT departments can let agents loose without losing the ability to see what they touched. Satya Nadella framed the whole package bluntly, telling the audience that Windows is no longer a platform for human users only, and that agents are now first-class citizens of the system.
Why This Matters More Than People Think
The 85 percent number is the headline, but the real maneuver is timing. The agent ecosystem is still forming, which means the developers who will define it have not yet picked a home. By offering nearly triple the take-home of the legacy app stores, Microsoft is making a bid to capture that talent during the brief window before defaults harden. Developer ecosystems are sticky in a way few markets are, because once an engineer learns a framework, ships on a store, and builds a customer base there, the switching cost becomes prohibitive. Microsoft learned this lesson painfully when it lost mobile, and it is determined not to lose agents the same way.
The deeper play is distribution, not economics. A generous revenue share is only valuable if there are buyers on the other side, and this is where Microsoft holds a hand no rival can match. Apple can offer an attractive cut but cannot put an agent inside the enterprise tools where knowledge work actually happens. Microsoft can place a third-party agent directly into Outlook, Teams, and the Windows taskbar of a 400 million-seat installed base that already runs payroll, email, and documents on its software. For a developer choosing where to build, the question is not who pays the best percentage, but who delivers the most users, and on that axis Microsoft's offer is close to unbeatable.
There is a defensive dimension too. OpenAI has its own ambitions for an agent marketplace, Salesforce is building AgentForce into a paid agent gate, and Google is wiring agents into Workspace. If any of them established the default registry for enterprise agents, Microsoft would find itself renting space on someone else's platform inside its own operating system. The Windows Agent Store is insurance against that future. By making Windows itself the runtime, the registry, and the storefront, Microsoft ensures that the agent economy, if it materializes, is built on Microsoft's rails and monetized through Microsoft's billing, no matter whose model is doing the thinking underneath.
Finally, the Agent 365 pricing reframes what Microsoft is actually selling. At 99 dollars per user per month, Microsoft is not monetizing the agents directly so much as creating a new per-seat subscription tier on top of the M365 base that hundreds of millions of workers already pay for. This is the same expansion-revenue motion that turned Office into a recurring cash machine, applied to a new category. The genius is that the 85 percent developer share costs Microsoft almost nothing if the real money comes from the seat license, because Microsoft captures the platform fee through the subscription while letting developers fight over the marketplace transactions on top.
The Competitive Landscape
The most direct targets are Apple and Google, whose 30 percent cut has been the subject of antitrust suits, the Epic Games trial, and years of developer fury. Microsoft is weaponizing that resentment, positioning itself as the platform that respects developers in exactly the way the mobile duopoly does not. But the more dangerous competitors sit in enterprise software. Salesforce's AgentForce and its AgentExchange marketplace, ServiceNow's agent fabric, and OpenAI's GPT and agent stores are all racing to become the place where companies discover and deploy AI agents, and each owns a slice of the enterprise workflow that Microsoft must now defend against.
The obvious historical parallel is Apple's 2008 App Store, which created the 30 percent standard and minted a generation of mobile fortunes. But a sharper analogy is Windows itself in the 1990s, when Microsoft won the desktop not by having the best operating system but by having the most software, because developers wrote for the platform with the most users and users bought the platform with the most software. That flywheel is exactly what Microsoft is trying to restart for agents, and the 85 percent share is the accelerant meant to get developers spinning it before a rival platform reaches critical mass.
The cautionary parallel is OpenAI's GPT Store, which launched with enormous fanfare in early 2024 and then struggled to produce either hit applications or real developer income. It demonstrated that building the storefront is the easy part and that generating real demand, discovery, and monetization is the hard part. Microsoft's answer is that it is not starting from a chatbot with a store bolted on, but from an operating system and a productivity suite that hundreds of millions of people are already required to use for work. Whether that distribution advantage converts into a thriving paid agent economy is the entire question, and it is far from settled.
A second instructive parallel is Steam, which won PC game distribution not on the size of its cut but on the depth of its tooling, its install base, and the network effect of being where the players already were. Valve charged a conventional fee yet became indispensable because it owned the relationship with the customer. Microsoft is attempting the same lock-in from the opposite direction, leading with an unusually generous split to seed the catalog, then relying on the Windows and M365 install base to make the store the place agents must be. The open question is whether agents behave like games and apps, where a central store became essential, or like web services, where developers ultimately preferred to own their own distribution and bypass any platform tax at all.
Hidden Insight: The Store Is a Trojan Horse for a New Subscription Tier
The framing everyone will adopt is platform-versus-platform, Microsoft against Apple in a fight over revenue share. That framing misses what Microsoft is actually doing, which is using a developer-friendly marketplace as the visible front of a far more lucrative move: a new top-tier enterprise subscription. The 85 percent share generates headlines and developer goodwill, but the 99-dollar Agent 365 seat is where the durable revenue lives. Microsoft has done this before, layering Copilot licenses on top of M365, and it works because enterprises buy seats, not transactions. The store is the marketing; the subscription is the business.
This explains why Microsoft can afford to be so generous on the split. A company that depended on marketplace fees would never give away 85 percent, because the platform cut would be its entire income. Microsoft does not need the cut, because it captures value at the seat level through the subscription that gates access to the agent ecosystem in the first place. The developer keeps 85 percent of a transaction that only exists because the customer is already paying Microsoft 99 dollars a month for the privilege of running agents at all. It is a toll booth disguised as a gift shop.
The bear case, however, is that none of this works without demand, and demand for paid agents is unproven. Skeptics point out that the GPT Store, the Salesforce AppExchange's early agent listings, and most enterprise AI pilots have struggled to show that companies will pay recurring fees for third-party agents rather than building their own or using the ones bundled free with their existing tools. The risk Microsoft is underpricing is that it builds a pristine marketplace with a beautiful revenue share and very few buyers, a mall with no shoppers, because the underlying behavior, paying a subscription per agent, may simply not take hold at the scale the 4.8 billion-dollar projection assumes.
There is a second, subtler risk in agent sprawl. The same frictionless registration that makes the Windows Agent Framework attractive to developers makes it a governance nightmare for the enterprises that buy it. An operating system where any agent can register a daemon, persist memory, and signal other agents over a communication bus is also an operating system with a vastly expanded attack surface and a new class of insider-threat vectors. AgentGuard exists precisely because Microsoft knows this, but security tooling shipped on day one rarely anticipates the creative misuse that emerges at scale. The companies most attracted to agent productivity are also the ones with the most to lose if an agent is compromised or simply misbehaves.
What to Watch Next
In the next 30 days, watch the developer signups and the early catalog. The 85 percent share is designed to produce a rush of registrations, and the raw count of agents listed, plus the quality of the launch partners beyond Adobe and Zoom, will tell you whether the incentive is working. Watch also for how OpenAI, Salesforce, and Google respond, because a matching or beating revenue share from a rival would signal that the platform war for agent developers has begun in earnest, and a price war on revenue splits would be a clear win for builders.
Over the next 90 days, the number that matters is paid attach rate. Microsoft will sell plenty of free or bundled agents, but the entire business case rests on enterprises buying the 99-dollar Agent 365 seats and on customers paying for premium third-party agents in the store. Watch the first earnings commentary that breaks out agent or Agent 365 revenue, and watch for case studies that show a real company paying real recurring money for a third-party agent rather than a free pilot. Without that, the marketplace is a demo, not a market.
On the 180-day horizon, the question is security and trust. The first serious incident involving a malicious or runaway agent registered through the framework will shape enterprise adoption more than any feature, because IT departments move on fear faster than on opportunity. Watch how AgentGuard performs under real-world conditions, watch for the first CISO guidance on agent governance, and watch whether Microsoft tightens or loosens the registration rules as it learns. The platform that wins the agent era will be the one enterprises trust to run agents safely, and that trust is earned in the incidents, not the keynotes.
Microsoft is giving developers 85 percent of a marketplace it built to sell something else entirely: a 99-dollar seat.
Key Takeaways
- 85 percent revenue share for developers, versus the 30 percent Apple and Google have charged for 17 years, is the store's central lure.
- 99 dollars per user per month for the Agent 365 seat is where Microsoft's durable revenue actually sits, not the marketplace cut.
- 4.8 billion dollars projected annually at just 10 percent adoption across 400 million M365 users, per Microsoft's own modeling.
- MIT-licensed Windows Agent Framework ships a registration daemon, a gRPC communication bus, and a persistent memory service for native agents.
- Adobe and Zoom launched as partners, with AgentGuard providing role-based access, data-loss prevention, and audit logging for enterprise control.
Questions Worth Asking
- If the real revenue comes from the 99-dollar seat rather than the marketplace cut, is the 85 percent developer share a genuine business model or a customer-acquisition subsidy that gets quietly trimmed once Microsoft wins the ecosystem?
- What evidence is there that enterprises will pay recurring fees for third-party agents at all, given how poorly the GPT Store and similar marketplaces have monetized?
- If any agent can register a daemon and signal other agents inside your operating system, how would your security team even know an agent had gone rogue before the damage was done?
