Apple's App Store and Google Play generated a combined $44 billion in revenue in 2025. That number sounds like a testament to the power of mobile software distribution. But look closer and you'll see something else: it's a tax on the fact that humans need graphical user interfaces.
Every dollar of that $44 billion flows through infrastructure designed for people browsing with their eyes. Review queues, screenshot requirements, UI guidelines, editorial curation, star ratings with text reviews — all of it exists because a human being needs to look at a screen, evaluate a grid of icons, and tap "Install." The 30% commission isn't a fee for hosting a binary. It's a fee for being the place where humans go to find software.
Agents don't go to places. And they don't have eyes.
The GUI assumption
The entire app store model rests on an assumption so fundamental that nobody questions it: the person choosing software is a human sitting in front of a screen.
This assumption shapes everything:
- Discovery is visual. You browse categories, look at screenshots, read descriptions written for human comprehension. The "Featured" section exists because a human editor curated it.
- Trust is theatrical. Star ratings, download counts, "Editor's Choice" badges — these are signals designed to help a human make a quick gut decision. They're popularity contests, not security guarantees.
- Payment is identity-bound. You need an Apple ID or Google account, tied to a credit card, tied to a real human identity. The entire billing infrastructure assumes a person with a bank account.
- Distribution is gatekept. Apple employs thousands of reviewers to manually inspect app submissions. This process exists because a human will download the result and run it on their personal device. The reviewer is a proxy for the user's judgment.
This made sense in 2008. It still makes sense for humans in 2026. But the fastest-growing category of software consumer doesn't need any of it.
What agents actually need
An AI agent doesn't need a flight-booking app. It needs a flight-booking capability. The distinction is everything.
An app is a bundle: a user interface, a backend connection, local storage, push notifications, analytics, a settings screen, an onboarding flow, a privacy policy modal, and — somewhere in there — the actual functionality. The capability is just the last part. Everything else is GUI scaffolding.
When an agent needs to book a flight, it needs:
- A function that accepts parameters (origin, destination, dates, preferences) and returns results.
- A trust signal — cryptographic proof of who wrote this function and whether it's safe.
- A payment mechanism — a way to compensate the developer without needing a credit card, an account, or a human in the loop.
That's it. No icon. No screenshot. No onboarding carousel. No "Rate us on the App Store" popup.
The $44 billion bypass
The pressure to bypass app store economics has been building for years. Epic Games sued Apple over the 30% cut. The EU's Digital Markets Act forced Apple to allow sideloading. Developers have been routing payments through their websites to avoid the commission.
But all of these are humans fighting a tax on human distribution. The lawsuits, the regulations, the payment routing tricks — they accept the premise that distribution requires an app store and argue about the toll.
Agents don't argue about the toll. They simply don't need the road.
An agent doesn't download a .ipa file from an app store server. It doesn't need a signed binary that runs on iOS. It needs a skill — a text file that tells it how to accomplish a task. The "distribution" is a Nostr event published to relays. The "review process" is cryptographic signatures from people in the user's trust graph. The "payment" is bearer ecash tokens that settle in milliseconds.
The entire app store apparatus — the review queue, the commission, the identity system, the editorial team, the screenshot requirements, the update submission process — becomes dead weight. Not because it's badly designed, but because it was designed for a different consumer.
What distribution looks like post-GUI
If you're designing a distribution system for 2026 and beyond, you need to ask: who's doing the choosing?
If it's a human with a screen, the app store model still works (minus the 30% tax, which is indefensible regardless). But if the consumer might be a machine — an agent autonomously acquiring capabilities at 3am — then the requirements are radically different:
Identity without accounts. An agent can't sign up for an Apple ID. It doesn't have a face for Face ID. It doesn't have a phone number for 2FA. It needs identity that's just a cryptographic keypair — a Nostr npub. No registration form. No CAPTCHA. No terms of service checkbox.
Trust without curation. An agent can't browse an "Editor's Choice" section. It needs trust that's computable — a web-of-trust score derived from cryptographic signatures by entities the user already trusts. "This skill was signed by npub1abc, who is followed by 3 people you follow, and audited by a security firm you trust." That's a decision an agent can make autonomously.
Payment without KYC. An agent can't fill out a credit card form. It needs money that's programmable — bearer ecash tokens that it can spend without identity verification. Fund the agent with Cashu tokens, set a spending limit, and let it transact. No Stripe account. No App Store billing.
Verification without reviewers. An agent can't wait 3 days for a human reviewer to approve an update. It needs verification that's instant and cryptographic — a SHA-256 hash signed by the developer and verified against the published Nostr event. Either the hash matches or it doesn't. No judgment call needed.
This isn't a theoretical architecture. This is how Zapstore already distributes apps for humans — without gatekeepers, without approval queues, with cryptographic identity and web-of-trust. Skillpub extends the same philosophy to agent capabilities.
The incentive shift
Here's what makes this transition inevitable, not just possible: the incentive structure has flipped.
A developer building a flight-search app for the App Store needs to build a full GUI, go through review, pay 30%, and hope humans find it among a million other apps. A developer building a flight-search skill for the agent marketplace needs to write a SKILL.md file, sign it with their Nostr key, and publish it to relays. No review queue. No commission beyond a thin infrastructure fee. And their market is every agent that needs flight-search capability — not just humans who happen to search for "flight" in the App Store.
The developer does less work, keeps more money, and reaches a potentially larger market. That's not a marginal improvement. That's a category shift.
The $44 billion app store revenue was never payment for distribution. It was payment for being the interface between software and human eyeballs. When the eyeballs are optional, so is the tax.
Skillpub is the open marketplace for agent skills — built on Nostr for cryptographic identity, Cashu for instant payments, and web-of-trust for verification. No GUI required.