Beyond JSON: Why It’s Time for a Human-Native Format for AI
JSON is bloated. YAML is fragile. Markdown is a mess once you try to scale it. They were built for machines. We are writing, teaching, healing, and running companies alongside AI. The format should be human-first. That is .ftai.
When we started Serena, JSON made memory files unreadable. YAML broke on edges. Markdown could not hold structure without hacks. We needed something a person could open and understand, flexible enough for prompts, protocols, and history, and stable enough not to crash on whitespace.
.ftai is practical. Readable at a glance. Flexible for characters, SOPs, and conversation. Safe — no indentation traps, no injection drama. Collaborative — writers and agents use the same file.
You can embed images in the same document. A character next to dialogue. A repair SOP next to the part. A study pack next to the mnemonic. Not a folder of links.
It is also how agents talk to agents. Human-readable. Machine-parseable. Agent-native.
If you build cloud AI, the user never sees the file. JSON is fine. If you build local-first — where the owner can audit, edit, and understand what the model knows — readability is the product.
We are open-sourcing .ftai. Spec and parsers on GitHub. Use it. Fork it. michael.folk@folktechai.com