The Post Apple Tax World | Agentic & Distributed Systems, DeFi, and Compute Economics

Apple took $10.1 billion in U.S. App Store commissions in 2024, double what they took in 2020. Then in one week of April 2025, the EU fined them €500M and a U.S. judge banned commissions on outside purchases. The native app pattern is being dismantled. Here's what comes next.

Apple took $10.1 billion in U.S. App Store commissions in 2024. That's double what they took in 2020.

Then in one week of April 2025, an EU regulator fined Apple €500 million for breaking the DMA and a U.S. federal judge banned them from charging any commission on purchases made outside the App Store. Two regulators on two continents, same conclusion, seven days apart.

The wall is coming down. A buddy of mine just shipped his golf app and asked me how to dodge the 30 percent cut. I told him: don't ship a native app at all.

Ship a web app. Then ship a skill. The agent does the input. The browser does the visuals. Apple never enters the picture.

He's not the first person to ask me this. He won't be the last. Everyone building a consumer product right now is staring at the same wall. The honest answer is you don't climb over it. You walk around it.

What Apple actually sells

Apple sells two things at the App Store. Most people only see one.

The first thing is distribution. A billion phones, a search bar, a featured tab. If you ship a native app, you can technically get in front of every iPhone user on earth. That's real.

The second thing is the user interface itself. Every iOS app is a tap-driven UI sitting on a 6-inch glass rectangle. You design for the rectangle. You ship for the rectangle. You collect taps.

Both of those services used to be the only game in town. If you wanted to build a consumer product in 2015, you built an iOS app. There was no other path that worked.

That's not true anymore.

The web caught up. Progressive web apps install to the home screen, look like native apps, run offline, send push notifications. The user can't tell the difference unless they actively look. And on iOS Safari, a PWA gives you about 90 percent of what a native app gives you for zero percent of the App Store tax.

That handles the distribution side. The other side is more interesting.

The interface stopped being the rectangle

Tapping a screen is the worst possible way to enter data when you have your hands on something else.

A golfer logging a round mid-game. A contractor logging a delivery. A driver logging an expense. Try doing any of those with a glass rectangle and ten taps. You won't. You'll forget. You'll log it later. You won't log it at all.

My buddy's golf app is the example. He wanted users to log birdies and eagles per hole per course. That's eighteen data entries per round, every round. Multiple courses. Some of them in country clubs where you're not pulling your phone out every hole.

The native app version of that is brutal. Open phone, unlock phone, find app, open round, find hole 7, tap birdie, save. Eighteen times.

The agent version is one sentence. "I shot a 78 at Innisbrook today, birdies on 7 and 14, eagle on 17." The agent parses it. The agent hits his API. The data lands.

That's not a small improvement. That's a different product.

The skill is the product

Here's the part most people miss when they hear "voice AI" and roll their eyes.

You don't need an agent platform to ship voice. You don't need a custom app. You don't need to convince users to download anything. You ship a SKILL.md.

A SKILL.md is a short instruction file. It tells any compatible agent how your product works. What API to call. What auth to use. What natural language to expect. What examples to follow. Anthropic shipped this as a literal product feature on October 16, 2025. It's not a thought experiment. It's a file format you can write today.

Any user with an agent installs the skill. Claude, ChatGPT, cYpher.camp, Gemini, whoever wins this round. The agent now knows how to talk to your product. Voice in, structured data out.

You shipped a SKILL.md. You didn't ship an app. You don't have an iOS team, you don't have a Play Store reviewer waiting, you don't have a 30 percent landlord. You have a markdown file and a public API.

The skill is the actual product. The web app is the dashboard.

Why this matters more than the Apple tax

When my buddy asked about the 30 percent, he was framing the problem as a cost problem. It's not.

It's a distribution problem disguised as a cost problem.

Apple's App Store is a search engine that takes a 30 percent cut. It worked because there was no other search engine for consumer apps. Now there is one. It's the agent.

ChatGPT hit 900 million weekly active users in February 2026, up from 500 million a year earlier. That's not a niche product anymore. That's a generation of users training themselves to ask an agent first and search second. If a user wants a golf tracker, they're going to ask their agent. The agent will install the right skill. The user will use the app. The App Store never enters the conversation.

This is the same shift that happened when Google replaced the Yellow Pages. The infrastructure that controlled discovery got replaced by infrastructure that answered intent directly. Skills are the next layer of that. Discovery used to be "open store, search, browse, install." It's becoming "tell agent what you want."

The 30 percent was never about the money. It was about being the gatekeeper of discovery. The gatekeeper is changing.

The new shape of a consumer app

The pattern I'm watching emerge looks like this.

You ship a web app. Mobile-responsive, PWA-installable, works on every device. That's your visual surface. Your dashboard. Your gallery. Your settings. The thing users open when they want to look at their data.

You ship a public API. Documented. Versioned. Auth via user-scoped keys generated in the web app. Standard REST or whatever's natural.

You ship a SKILL.md. Free to download. It teaches any compatible agent how to call your API on the user's behalf. This is what users connect their agent to when they want fast input or voice-driven workflows.

You wrap it with three tiers. Free tier gets the web app. Paid tier unlocks the skill. Higher tier unlocks deeper agent capabilities. Peer benchmarking, autonomous workflows, scheduled actions.

There's no iOS app in this shape. There's no Play Store either. There's no need for one.

I'm already shipping this

This isn't a prediction. I'm running it.

cYpher.camp is a web app, a public API, and a SKILL.md. Paying customers. Real revenue. No iOS team, no Play Store reviewer, no 30 percent landlord. The exact pattern this post describes is live in production today.

My buddy's not building on cYpher.camp. He's building a SKILL.md that any agent runtime can load. If cYpher.camp dies tomorrow, his skill works on Claude. If Claude dies, his skill works on whatever comes next. The skill is portable infrastructure. The agent is interchangeable.

What he is doing is putting a cYpher.camp referral link on his website. Users who don't have a voice agent need one to use the skill. He sends them through cYpher.camp. He gets a referral credit. cYpher.camp gets a high-LTV user. The user gets an agent.

Everyone wins. Nobody is locked in.

You don't build on top of a platform. You build infrastructure that any platform can pick up. The platform becomes a discovery channel, not a dependency.

The four-step playbook

Ship a PWA on the web first. Get to a working product without the App Store overhead. You can wrap a native shell later if the unit economics demand it. Most won't.

Build an API from day one. Even if your only client is your own web app, design it like other clients will hit it. Document it. Version it. Auth with user-scoped keys.

Ship a SKILL.md when you have your first paying users. Pick one agent runtime to test against, support it well, document the rest as community-supported.

Position the skill as the experience, not the technology. Don't say "we have an AI integration." Say "log your round by voice." The user doesn't care that you wrote a markdown file. They care that they don't have to tap.

The honest counterpoint

This is not free.

Agent runtimes are still early. SKILL.md as a standard is still being defined. Half the agent platforms in 2026 won't exist in 2028. The user experience varies wildly between Claude and ChatGPT and the rest. You will get bug reports about your skill working in one runtime and not another.

Not every user wants to talk to their phone. Plenty will. Voice is going to be the input layer for an entire generation of products, but it's not going to be the input layer for every product. Some workflows are better with taps. Some users won't adopt voice no matter how much smoother it is.

The web-first approach gives up less than people think. Apple Watch? Your users dictate to Telegram on their wrist and the agent does the rest. iOS push notifications? PWA web push shipped on iOS Safari in March 2023, so even that gap closed three years ago. The remaining gaps are tight camera integration and native game-engine performance. If your product genuinely needs those, native still matters. For 95 percent of consumer apps, it doesn't.

One caveat I owe you. The April 30 ruling that banned Apple from charging commissions on outside purchases was partially walked back by the 9th Circuit in December 2025. Apple appealed to the Supreme Court in April 2026. The framework for how this plays out long-term is still being written. But the trend is set. Regulators are no longer pretending Apple's gate is reasonable, and developers are no longer pretending they have no other path.

I'm not saying you should never ship native. I'm saying the default has flipped. Native used to be "of course." Now it's "only if you have a specific reason." The web plus a skill is the default starting point. Native is the optimization you ship later if the use case demands it.

The bigger picture

Every platform shift has the same shape. The dominant interface paradigm gets replaced. The companies that hard-couple to the old paradigm lose. The companies that build at the right abstraction layer win.

Web replaced desktop. Mobile replaced web for some things, then the web caught up. Now agents are replacing mobile for input-heavy use cases. Mobile will still matter. Your dashboard is on a phone. Your tap surface is on a phone. But the act of inputting data is moving to voice.

If you're building a consumer product in 2026 and your entire roadmap depends on an iOS launch, you're building for the previous era. The 30 percent tax is the loudest signal. The bigger signal is that you're locking yourself into an interface paradigm that's about to get less important.

The agent is the new search bar. Apple just stopped being the gatekeeper.