“Beyond JSON: Why It’s Time for a Human-Native Format for AI”
- Michael Folk
- Dec 5, 2025
- 3 min read
Updated: Feb 14
Most people don’t think twice about how computers “talk.” But if you’ve ever copied a chatbot prompt, wrangled a config file, or stared down a wall of curly braces, you’ve felt it — the alien language of machines.
Remember the movie The Good, The Bad, and The Ugly? That’s exactly what it’s like dealing with today’s formats.
“That new kid’s got range — and it speaks human.”
JSON is bloated.
YAML is fragile.
Markdown is a mess once you try to scale it.
They’ve all had their moment — but in 2025, they’re showing their age.
These formats were never meant for people. They were built for machines — and they suck when used outside that world.
But AI has changed everything. We’re not just coding with it anymore — we’re writing, teaching, healing, building businesses, even raising kids alongside AI. And the formats we use should evolve with us.
We don’t need another machine-readable syntax.
We need a format that’s human-first.
That’s what .ftai is.
What Makes .ftai different from Json?
.ftai isn't revolutionary — it's practical.
When we started building Serena (our AI assistant), we hit a wall: JSON made memory files unreadable. YAML broke on edge cases. Markdown couldn't handle structured data without hacks.
We needed a format that was:
Readable by humans — you shouldn't need a parser to understand what your AI remembers
Flexible enough for AI — prompts, protocols, character files, conversation history
Stable and safe — no indentation crashes, no injection risks, no surprises
So we built .ftai. Not to replace JSON everywhere — just to solve the problems JSON creates when humans need to actually read and edit AI data.
Human-readable. Machine-parseable. Agent-native.
🖼️ You can embed images inline — in the same format.
Not Markdown hacks. Not separate folders. Just direct visual context.
AI can see what you see, right where you wrote it.
Think:
A character profile with an image next to the dialogue.
A repair SOP with inline part diagrams.
A study pack with flashcards and visual mnemonics — side-by-side.
No other format gives you that level of contextual clarity.
Not JSON. Not YAML. Not Markdown. .ftai does.
Here’s what else sets .ftai apart:
Readable at a glance. You don’t need to “parse” .ftai — you just read it.
Flexible by design. Prompts, protocols, characters, SOPs — .ftai handles it.
Safe and stable. No indentation crashes. No injection risks. No format drama.
Collaborative. Writers, engineers, teachers, agents — all using the same format.
This isn’t just another markup flavor.
It’s the missing link between humans and the AI tools we actually use.
What .ftai Actually Does
1. Human-Readable AI Memory
When Serena stores a conversation or protocol, it's saved in .ftai. You can open the file and actually read it. No curly braces. No escape characters. Just structured text that makes sense.
2. Inline Context (Including Images)
.ftai lets you embed images directly in the same file — not as links or separate folders, but as inline visual context.
Think:
A character profile with an image next to the dialogue
A repair SOP with inline part diagrams
A study pack with flashcards and visual mnemonics side-by-side
No other format gives you that level of contextual clarity without hacks.
3. AI-to-AI Communication
Here's the twist: .ftai isn't just for humans.
When AI agents need to coordinate — pass data between tools, share protocols, maintain consistency — .ftai standardizes how they talk. No more mismatched prompts or fragile JSON pipelines.
Human-readable. Machine-parseable. Agent-native.
Why This Matters (Especially for Local-First AI)
If you're building cloud-dependent AI, JSON is fine. The user never sees the data anyway.
But if you're building local-first AI — where the user owns their data, can audit it, modify it, and understand what the AI knows — then readability isn't optional. It's foundational.
.ftai is how we make Serena transparent by design. You can:
Open Serena's memory files and read them
Edit protocols without breaking syntax
Share character files with other users
Audit exactly what Serena knows about you
This isn't just a technical decision. It's an ethical one.
🛠️ So what comes next?
We're open sourcing .ftai.
If you're building local-first AI and tired of wrestling with JSON, YAML, or Markdown hacks — use it. Fork it. Improve it.
The spec and parsers will be on GitHub. We use .ftai internally for Serena and other systems we're building, but we're not gatekeeping it.
If you're interested in .ftai or want to talk about human-readable AI formats, reach out.
Start writing like a human again.
Comments