The JSONification of Everything

More and more of the world is structured as JSON (or equivalent: YAML, structured logs, API payloads). That’s not new, but in 2025–2026 the trend has accelerated: LLMs natively produce and consume JSON; agents pass state as JSON; configs, prompts, and tool schemas are JSON. The "JSONification of everything" is the observation that once something is in a consistent schema, it can be generated, validated, transformed, and piped by both humans and models.

Implications:

• APIs and tool use. Tool-calling and function-use in LLMs are schema-driven (e.g. OpenAPI, JSON Schema). The more your systems speak JSON with clear schemas, the more they plug into agent workflows and automation.

• Content and config. CMS entries, feature flags, and prompt templates stored as structured data are easier to version, A/B test, and hand to models for editing or generation.

• Observability and evaluation. Logs, traces, and eval datasets as JSON (or similar) make it straightforward to analyze, replay, and improve pipelines and agents.

The downside is the same as with any standard: everything starts to look like a nail. Not every problem is best modeled as nested key-value structures. But for integration, automation, and AI-native pipelines, JSONification is the dominant trend—and designing your data and APIs with that in mind pays off.