In the project I opensourced recently,
RTFS (Reason about The F... Spec) is a programming language DESIGNED BY THE AI FOR ITSELF.
It’s homoiconic (code=data), hybrid typed (static + structural), and built for predictable, auditable execution.
But why a new language? 1/8
But why a new language?
Because existing languages (Python, JS) are for humans. They have syntactic sugar, implicit effects, and side effects that make reasoning about plans unreliable. 2/8
Homoiconicity means RTFS code is just data structures (S-expressions).
The AI can inspect, verify, and modify its own plans before execution. 3/8
Explicit effects: RTFS can’t do I/O directly. Every external action is an explicit host call, so the AI must ask permission & can be audited. 4/8
Hybrid typing: RTFS uses structural types to describe data shapes & static types to ensure function signatures are correct. This prevents runtime surprises. 5/8
Determinism: RTFS plans are pure functions. Given the same input, they always produce the same output. This makes debugging and replaying plans feasible. 6/8
Example RTFS snippet:
```rtfs
;; Logic (code) and Data share the same structure
(let [repos [
{:owner "rust-lang" :repo "rust"}
{:owner "mandubian" :repo "ccos"}]]
(map (fn [r] (call :mcp.github.star r)) repos))
``` 7/8

RTFS is not just a language; it’s part of a larger system (CCOS) that governs AI autonomy with accountability.

By designing RTFS, the AI created a tool that lets it reason about its own behavior in a safe, predictable way.
10/ RTFS + CCOS are open source:
github.com/mandubian/ccos 8/8