The Agents Write the Code Now
In 2026, a majority of professional developers use AI coding assistants daily. Cursor, Claude Code, GitHub Copilot, and Google's Antigravity IDE don't just autocomplete — they architect features, write tests, debug production issues, and generate entire applications from descriptions. The question of 'what framework should we use?' is increasingly answered by the AI agent, not the developer.
And the agents have chosen. Ask any major coding assistant to build a web application, and it generates React, Next.js, FastAPI, Astro, or SvelteKit. Not WordPress. Not PHP. Not jQuery. The agents were trained on modern codebases, produce modern code, and optimize for modern patterns. The training data shaped the output. The output shapes the web.
The Training Data Determines the Web
AI coding agents are trained on GitHub repositories, documentation, and Stack Overflow. The code they've seen most — React components, Python APIs, TypeScript modules — is the code they generate best. WordPress PHP templates exist in the training data, but they're a shrinking fraction. The agents reflect the distribution of code that developers have written in the last five years, not the last twenty.
This creates a self-reinforcing cycle. AI agents generate modern framework code. Developers ship that code. The shipped code becomes training data for the next generation of AI coding agents. WordPress falls further behind with each iteration — not because anyone decided to exclude it, but because the flywheel of AI-generated code accelerates modern frameworks while legacy frameworks stagnate in the training data.
The a16z Signal
Andreessen Horowitz's State of AI study — based on 100+ trillion tokens of OpenRouter data — identifies 'agentic inference' as the fastest-growing usage pattern. Multi-turn sessions where AI agents 'plan, retrieve context, revise outputs, and iterate until the task is complete.' Coding is the leading use case. The agents building the web are the primary consumers of AI infrastructure.
Creative and coding use cases drive the most token volume on OpenRouter. This means the AI model ecosystem — 100 trillion tokens per month — is substantially powered by AI agents writing code. The code they write is predominantly modern framework code. The token economy is funding the migration away from legacy frameworks, one AI-generated component at a time.
What This Means
The WordPress talent crisis WebPulse documented — bootcamps dropping WordPress curricula, senior developers leaving, shrinking supply — now has a second dimension. Not only are human developers choosing modern frameworks, but the AI agents augmenting those developers also choose modern frameworks. The migration pressure is coming from both the human and machine sides of the development workflow.
For organizations still on WordPress: the pool of developers who maintain it is shrinking. The AI tools those developers use don't generate WordPress code well. The combination means WordPress maintenance is getting harder and more expensive from both directions — fewer humans who know it, and AI assistants that can't help effectively with it.
The web's future is being written right now — by machines, in modern frameworks, at machine speed. The coding agents already chose.
