When Apps Cost Nothing, Protocols Are Everything

AI coding tools are driving the cost of building an app toward zero. When everyone can build an app, apps stop being scarce. What becomes scarce — and valuable — is the shared understanding between apps. That shared understanding is a protocol.

This isn’t speculation. I use Claude Code to write 100% of my code. Mitchell Hashimoto — the creator of Terraform — has restructured his entire workflow around AI agents. David Heinemeier Hansson — creator of Ruby on Rails — describes the same shift. The cost collapse is measurable, month over month, model over model.

When everyone has their own app, how do they all talk to each other? The answer is protocols.

Protocols Outlive Apps

A protocol is an agreement about how to communicate. TCP/IP, HTTP, SMTP, Bitcoin, HTML, JSON — all protocols. File formats are protocols too. A PDF is an agreement between writer and reader. A JPEG is an agreement about how an image is compressed.

The key property: protocols are permissionless. Anyone can build an email client, a web browser, a Bitcoin wallet. No API key, no terms of service. You just implement the protocol.

And protocols outlive every app built on top of them. SMTP is 44 years old. Every email client ever built — Eudora, Outlook Express, Thunderbird, Sparrow, Inbox by Google — is dead or marginalized. SMTP is stronger than ever. HTTP is 35. Mosaic, Netscape, IE6 — all gone. HTTP is the backbone of the modern economy. Bitcoin is 17. Mt. Gox, BitInstant, BTC-e — all gone. The Bitcoin protocol processes billions daily.

Protocols are the bedrock. Apps are the topsoil.

The Age of the Mega-App

From roughly 2004 to 2024, the dominant pattern was the mega-app: a single application used by billions, controlled by a single company. Facebook, YouTube, WhatsApp, Uber, Slack, TikTok, Salesforce. Winner-take-all dynamics driven by network effects. Everyone uses Facebook because everyone is on Facebook.

This created massive data silos. Your social graph locked in Facebook. Your messages locked in WhatsApp. Your files locked in Dropbox. Interoperability is against these companies’ business interests. Your data is their moat.

This was the only viable model when building an app was expensive. When it costs millions and years to build a competitive social network, only a few survive. The barrier to entry was the moat.

That barrier is collapsing.

AI Is Collapsing the Cost of Apps

The history of dev tools is a history of transferring labor from human to machine. IDEs automated compilation. IntelliSense automated lookup. Copilot automated boilerplate. In 2025, agentic AI — Claude Code, OpenAI Codex, and others — automated implementation itself. Agents plan, write, run, test, debug, and iterate autonomously. They close their own feedback loop.

This is different in kind from everything before. An IDE could show you the error; it couldn’t fix it and try again. Copilot could suggest a line; it couldn’t reason about your architecture across 200 files. Agents do both.

I spend $500–1000/month on AI agents. A junior engineer costs $8k–15k/month. The agent is faster, works weekends, and scales with the number of tasks I can define. The economics are irrefutable. And the tools improve on weekly release cycles. The cost of building an app is in free fall.

When Apps Are Cheap, Protocols Become Valuable

If anyone can build an app, the number and diversity of apps explodes. When building a social media app takes days instead of years and dollars instead of millions, the mega-app model breaks down. Why use someone else’s app — with their design choices, their algorithmic feed, their data policies — when you can have your own?

But humans still need to collaborate. The value was never in the app — it was in the connection to other people and their data. The mega-app solved this by putting everyone on one platform. That was a centralization hack, not an ideal solution.

The ideal solution already exists for email and the web. You don’t use the same email client as everyone else — Gmail, Outlook, Thunderbird, mutt — they all speak SMTP and IMAP. You don’t use the same browser — Chrome, Firefox, Safari — they all speak HTTP and render HTML. The protocol is the shared layer; the app is your choice.

This model is about to generalize to everything. Social media, messaging, project management, ride-sharing, e-commerce — each served by open protocols with diverse, customizable apps on top. ActivityPub, AT Protocol, Nostr, Matrix, and MCP are early attempts. The direction is clear.

The New Software Engineer: Protocol Designer

If AI agents can build apps, the high-value work shifts to protocol design. What data formats should we standardize? How do we handle identity across apps — cryptographic key pairs instead of platform accounts? How do apps exchange value — Bitcoin and Lightning as open payment protocols? How do AI agents coordinate — MCP as a standard interface? What’s the equivalent of HTML for a recipe, a medical record, a ride request?

Designing a protocol is harder than building an app. Protocols must be simple enough to implement, flexible enough to evolve, secure enough to trust, and compelling enough to adopt. AI agents can write a CRUD app in minutes. They cannot design a protocol that thousands of teams will adopt for decades.

Building a successful protocol means building a reference app — but the app is the proof-of-concept, not the product. Satoshi built Bitcoin Core to prove the Bitcoin protocol. Berners-Lee built the first browser to prove HTTP and HTML. The protocol is the product.

The cost of building an app is approaching zero. The value of a good protocol is approaching infinity. The age of the mega-app is ending. The age of the protocol is beginning.