A programming language with no compiler.
"Human language is the best programming language in the future." — Jensen Huang
humanlang is a programming language where the syntax is natural English. There is no compiler. There is no interpreter. There is only a specification, a set of tests, and you — talking to an AI agent that builds the implementation.
-- FizzBuzz in humanlang
for n from 1 to 100:
if n modulo 15 is equal to 0:
print "FizzBuzz"
otherwise if n modulo 3 is equal to 0:
print "Fizz"
otherwise if n modulo 5 is equal to 0:
print "Buzz"
otherwise:
print n
-- Functions and lists
define greet with name:
return "Hello, " joined with name joined with "!"
set guests to ["Alice", "Bob", "Charlie"]
for each guest in guests:
print call greet with guest
-- Recursion
define factorial with n:
if n is at most 1:
return 1
return n times (call factorial with (n minus 1))
print call factorial with 10
| File | What It Is |
|---|---|
| SPEC.md | Complete language specification with EBNF grammar |
| tests.yaml | 86 language-agnostic test cases |
| INSTALL.md | Instructions for building humanlang (a prompt) |
| examples/ | Example programs in .hl format |
| python/ | AI-generated Python interpreter (1,465 lines) |
| typescript/ | AI-generated TypeScript interpreter (2,188 lines) |
| kotlin/ | AI-generated Kotlin interpreter (1,332 lines) |
| rust/ | AI-generated Rust interpreter (2,103 lines) |
| ocaml/ | AI-generated OCaml interpreter (1,429 lines) |
| zig/ | AI-generated Zig interpreter (1,741 lines) |
| haskell/ | AI-generated Haskell interpreter (1,243 lines) |
| asm/ | AI-generated ARM64 assembly interpreter (5,271 lines) |
AI-generated interpreters, each built from nothing but SPEC.md and tests.yaml. Different paradigms. One spec. All passing 86/86 tests.
| Language | Paradigm | Lines | Files | Build |
|---|---|---|---|---|
| Python | Dynamic / Scripting | 1,465 | Single file | python3 humanlang.py |
| TypeScript | Typed JS / Node.js | 2,188 | Single file | npx tsx humanlang.ts |
| Kotlin | JVM / OOP+FP | 1,332 | Single file | kotlinc humanlang.kt -include-runtime -d humanlang.jar |
| Rust | Systems / Ownership | 2,103 | 6 modules | cargo build --release |
| OCaml | Functional / ADTs | 1,429 | Single file | ocamlopt -o humanlang humanlang.ml |
| Zig | Manual memory / Zero overhead | 1,741 | Single file | zig build -Doptimize=ReleaseSafe |
| Haskell | Pure lazy functional | 1,243 | Single file | ghc -O2 -o humanlang Main.hs |
| ARM64 Assembly | Raw machine instructions | 5,271 | Single file | cc -o humanlang humanlang.s |
# Python
cd python && python3 humanlang.py ../examples/fizzbuzz.hl
# TypeScript
cd typescript && npm install && npx tsx humanlang.ts ../examples/fizzbuzz.hl
# Kotlin
cd kotlin && kotlinc humanlang.kt -include-runtime -d humanlang.jar
java -jar humanlang.jar ../examples/fizzbuzz.hl
# Rust
cd rust && cargo build --release
./target/release/humanlang ../examples/fizzbuzz.hl
# OCaml
cd ocaml && ocamlopt -o humanlang humanlang.ml
./humanlang ../examples/fizzbuzz.hl
# Zig
cd zig && zig build -Doptimize=ReleaseSafe
./zig-out/bin/humanlang ../examples/fizzbuzz.hl
# Haskell
cd haskell && ghc -O2 -o humanlang Main.hs
./humanlang ../examples/fizzbuzz.hl
# ARM64 Assembly
cd asm && cc -o humanlang humanlang.s
./humanlang ../examples/fizzbuzz.hlEach implementation includes a run_tests.sh that validates all 86 test cases:
cd python && bash run_tests.sh # 86/86 ✓
cd typescript && bash run_tests.sh # 86/86 ✓
cd kotlin && bash run_tests.sh # 86/86 ✓
cd rust && bash run_tests.sh # 86/86 ✓
cd ocaml && bash run_tests.sh # 86/86 ✓
cd zig && bash run_tests.sh # 86/86 ✓
cd haskell && bash run_tests.sh # 86/86 ✓
cd asm && bash run_tests.sh # 86/86 ✓No implementation was hand-written. Each was generated by an AI agent reading the spec and iterating until all tests passed. The spec IS the compiler.
Give your AI coding agent this prompt:
Build a humanlang interpreter in [LANGUAGE].
1. Read SPEC.md for the complete language specification
2. Parse tests.yaml and generate a test file
3. Implement a lexer, parser, and tree-walk interpreter
4. The entry point should accept a .hl file path as a command-line argument
5. Run tests until all pass
6. Place implementation in [LOCATION]
Pick your language. Pick your location. Copy, paste, go.
humanlang supports:
- Variables —
set x to 10 - Arithmetic —
plus,minus,times,divided by,modulo - Strings —
joined with,uppercase of,lowercase of,contains - Comparisons —
is equal to,is greater than,is at most, etc. - Logic —
and,or,not - Conditionals —
if/otherwise if/otherwise - Loops —
repeat N times,for each,for N from X to Y,while - Functions —
define,call,return, recursion - Lists —
first of,last of,append,item N of,contains - Type conversion —
as number,as string,as boolean
See SPEC.md for the complete grammar and semantics.
Ghost libraries — software distributed as specifications rather than code — are gaining traction. A spec and tests that any AI agent can implement in any language. But a library is small.
What if a spec could replace a compiler?
The answer is yes.
humanlang is a complete programming language — variables, functions, loops, recursion, lists — that exists entirely as a specification. No binary. No runtime. No dependency. Just a document that describes how English becomes computation.
The spec IS the compiler. The AI agent IS the build tool. Human language IS the programming language.
MIT