A live catalogue of an emerging field. 38 projects, three camps, and a foundational argument about how to frame the problem itself.
Originally catalogued in a post “Three camps alike in dignity” written for the Negroni Venture Studios blog.
The Syntactic camp says the problem is representational — strip ambiguity at the token level. The Verification camp says it's semantic — make contracts mechanically checkable. The Orchestration camp says it isn't a language problem at all — constrain how agents coordinate. The catalogue below treats them as evidence that this disagreement is real.
If the problem is that models trip on syntax, the fix is to strip ambiguity from the syntax itself.
The syntactic camp treats the problem as one of representation. Models choke on tokens that mean different things in different positions, on operators that need disambiguation, on whitespace that might or might not be load-bearing. Their answer: build a syntax where every token has one job.
The strongest entries replace text with structure (X07's JSON ASTs), eliminate operators in favour of keywords (NERD), or surface intermediate representations as the user-facing form (Magpie's SSA). The weakest are exercises in extreme density — one-character opcodes — that read more as conceptual art than as production languages.
What unites them is a belief that the LLM's job is easier if the surface is simpler. What divides them from the verification camp is the absence of any mechanism for the compiler to catch what the model gets wrong.
The model doesn't need to be right. It needs to be checkable.
The verification camp accepts that LLMs will keep making semantic errors and asks a different question: can the compiler catch them? Their answer is mandatory contracts, refinement types, effect systems, and SMT-backed proofs — the machinery of formal methods, repurposed as a guardrail for generative code.
This is the camp with the most established theoretical foundation (the work goes back to ML, Coq, Dafny, and Lean) and also the most mature implementations. MoonBit ships a full toolchain with ICSE-published research; Vera ships measured benchmark wins against Python on zero training data; AILANG claims to be written autonomously by its own coordinator.
The provocative entry is Prove — same diagnosis, opposite conclusion. It uses verification to make the language resistant to AI generation. The licence explicitly prohibits training use. Same camp, opposite politics.
It isn't a language problem. It's an agent-coordination problem.
The orchestration camp re-frames the question. The trouble with LLM-authored code, they argue, isn't any specific defect in the code — it's that agents need to be sequenced, sandboxed, audited, and approved by humans at the right points. The language is just the substrate; the runtime is where the action is.
Some entries here are academic (Pel from CMU, Quasar from Penn); some are infrastructure (Marsha treats the LLM as the compiler itself); one is a serious engineering effort aimed squarely at regulated industries (Boruna's hash-chained evidence bundles and deterministic replay).
This camp overlaps the most with the others — Boruna includes a type checker and capability system; Quasar adds conformal prediction. The line between “language for agents” and “framework for agents” is genuinely blurry here, which is itself informative.
Every language designed for LLMs or agents to author code, as of 17 aug 2026.
Backend API language for small LLMs (1B/3B/7B). Twelve top-level constructs cover a production backend; LL(1) grammar and prefix-only expressions ship per-state logit masks for constrained decoding.
Rust · Native artefacts (SQL DDL, Rust/axum server, TypeScript and Rust client SDKs, OpenAPI, GraphQL); also interpreted directly from the AST via axum + sqlx
A Jason Hall blog post on three attempts at an LLM-optimised language: B-IR with unicode opcodes, TBIR with control characters the model rewrote into English keywords, and Loom with pre/postconditions and stable error codes.
Python (bootstrap) · Arm64 assembly (Mach-O via clang)
Designed for AI to write, with one canonical way to express every operation. Nine bundled modules, structured JSON errors with fix/retry fields, ? operator for propagation. Compiles to Go IR, then native via go build.
Go · Native binary via Go IR + `go build`
Token-minimal language where every builtin has a short fixed alias and design changes are argued against measured token cost. Development is driven by scripted agent sessions whose failures are filed as language bugs.
Rust · Native (Cranelift JIT and object emission), with a bytecode VM
Minimal indentation-based syntax with no punctuation. A Python bootstrap compiler emits C in memory and pipes it to cc -O2. Framed by its author as a weekend experiment in what an LLM produces when asked to design a language for itself.
Python (bootstrap) · C (via gcc/clang)
Prefix-arity AST with single-character ASCII operators and De Bruijn variable indices. Linear ownership enforced at compile time. Compiler-injected OpenTelemetry spans triggered by a metadata marker. LLVM IR via Rust, OpenCL JIT for GPU map operations.
Rust · LLVM IR (then native via clang); OpenCL JIT for GPU map kernels at runtime
AI-first backend language, immutable by default, with one canonical way to express each operation. Ships a built-in token-budgeted retrieval tool (lume kb) that packs local language docs, examples, and structured diagnostics under a caller-set token cap.
Go · Native binary via Go transpilation + `go build`
SSA as the surface syntax. Every value %-prefixed and typed at definition; one canonical way to express each operation; compiles to native via LLVM.
Rust · LLVM IR / native, also WebAssembly
Statically typed embedded language with flat operators (no precedence) and host-granted capabilities. Full spec fits in 3,200 tokens. Compiles to native via a safe-Rust port of QBE.
Rust · Native (via rqbe, an in-process safe-Rust port of QBE)
No Effort Required, Done. Replaces every operator with an English keyword on the bet that LLM tokenisers spend fewer tokens on words than on symbols. Built-in MCP client and LLM call primitives.
C · LLVM IR (then native via clang)
Single-character opcodes for extreme density. The author's README disclaims the entire repository as Claude-generated and explicitly frames the project as a thought experiment or art piece.
Zig (Claude-generated) · Native (via Zig backend, claimed)
AST as the source of truth. Canonical byte-exact text, BLAKE3-addressed definitions, DeBruijn indices, typed Hole nodes for malformed code, and explicit effects in function signatures.
Rust · LLVM IR / native (x86_64 Linux)
Eliminates text syntax. Programs are canonical JSON ASTs (x07AST); edits are RFC 6902 JSON Patch operations; diagnostics ship as stable JSON with quickfixes the toolchain applies deterministically.
Rust · Native (via C codegen); WebAssembly
Row-polymorphic Hindley-Milner with capability-based effects (IO, FS, Net, Clock, AI). No loops, lambda calculus only. Written autonomously by AI agents.
Go · Native binaries, WebAssembly
Every function carries intent, declared effects, and a colocated verify block. Pure verify blocks export to Lean 4 theorems and Dafny lemmas; effectful ones lift through the Oracle proof export.
Rust · bytecode VM, Rust, WebAssembly GC + wasip2 (Lean 4 / Dafny via proof export)
Not a new language. The bet is that Haskell's purity and semantic density already make AI-written, verifiable compute feel natural — once toolchain friction is removed. hx wraps cabal/stack/ghcup/HLS in Rust; BHC is a clean-slate Haskell 2026 compiler with per-profile runtimes.
Rust · GHC/Cabal/HLS (hx, wrapping); LLVM, WebAssembly, GPU (BHC, in development)
Self-hosting literate language on bare-metal x86-64: dependent, linear, and effect types, proof blocks, hard real-time enforcement, and a byte-identical fixed point as the acceptance gate for every change. Authored and maintained end-to-end by AI agents.
Codex (self-hosted; the bootstrap reference compiler is retired) · Bare-metal x86-64 (CDX binary, no OS or libc); ARM64 and RISC-V cross-compilation via plugs; PTX (NVIDIA) and SPIR-V (Vulkan/OpenCL) GPU targets; 53 transpiler plugs total including WebAssembly
Concurrent systems language where architecture is mechanically checked: compile-time effect certificates and reachability claims that fail closed on unresolvable indirection. Human-and-LLM authorship is one of four stated design pillars.
Rust · Native via LLVM; WebAssembly
Mandatory preconditions, postconditions, and entity invariants. Z3 SMT verification via intentc verify. Natural-language intent blocks that resolve to specific contract references. One source file compiles to Rust, JavaScript, and WebAssembly.
Go · Native binaries (via Rust), JavaScript, WebAssembly (direct binary)
AI-friendly general-purpose language. ICSE 2024 paper on real-time semantics-aware token sampling. Three years of training data.
OCaml · WASM GC, JavaScript, native (C codegen), LLVM
Mandatory shadow test blocks on every function. The proved core (NanoCore) has 193 Coq theorems with zero axioms. Multi-target codegen across C, WebAssembly, LLVM IR, PTX, and RISC-V.
C · C (default), WebAssembly, LLVM IR, PTX, RISC-V
Intent blocks on every function and route, pipeline syntax, explicit effects, errors as types. Single binary that ships an HTTP server, SQLite, an LSP, and an MCP server with five tools.
Rust · Interpreted (tree-walking)
Intent-first language with verb-based IO, refinement types, and contracts. Source is covered by the Prove Source License v1.0, which prohibits use as AI training data.
Python (bootstrap) · C (then native via gcc/clang)
Contract-first language with mandatory req, ens, and fx clauses; Forge reports per-obligation assurance through Verus, bounded checking, runtime contracts, or Lean-checked reconstruction.
Rust and Lean 4 · Rust; native Linux executables and freestanding libraries via rustc
Mandatory contracts on every function. Z3 SMT verification with a runtime-check fallback. Typed slot references replace variable names. LLM inference is a first-class typed effect.
Python · WebAssembly (core; browser bundle; experimental WASI Preview 2 component, including a wasi:http server world)
Every function carries machine-checked vows. ESBMC bounded model checking discharges them at compile time. The compiler binary ships its own Claude Code skill, generated from the same source as the toolchain.
Rust (stage 0); self-hosted in Vow · Native (via Cranelift); C (for the ESBMC verification pipeline only)
Vercel Labs' agent-first systems language. Sub-10 KiB native binaries. Structured JSON diagnostics with stable codes and typed repair plans. One obvious path.
C (zero-c bootstrap); self-hosted compiler-zero in progress · Native binaries (direct ELF/Mach-O/PE emitters, no LLVM), WebAssembly
Deterministic, capability-safe workflow execution. Every effect declared, policy-gated. Hash-chained tamper-evident evidence bundles.
Rust · Bytecode (custom VM)
Workflow harness for coding agents. .fabro files are Graphviz digraphs whose node shapes pick handlers; a CSS-like stylesheet routes each step to a model. Single Rust binary.
Rust · Native binary (Rust workspace, embedded React SPA)
Markdown-native source (.lm.md). Algebraic effects, grants for tool and model calls, @deterministic compile-time enforcement, and pipeline / machine / memory process kinds. A language for humans authoring agent workflows.
Rust · LIR bytecode → register-based VM (~100 opcodes); WebAssembly via lumen-wasm
Functional, English-based language whose .mrsh files (declaration, description, examples) are compiled to tested Python by an LLM. Alpha implementation; last maintainer activity early Aug 2023.
Python · Python (generated by an LLM, with auto-generated tests)
Lisp-flavoured language for orchestrating LLM agents, with capability control enforced at the grammar level and a REPeL self-healing loop modelled on Common Lisp restarts.
N/A (paper-only) · N/A (paper-only)
Catalog-hosted path expressions over typed API graphs, with session-scoped opaque symbols, dry-run execution plans, and federated multi-API sessions.
Rust · Native binaries (HTTP/API execution plans)
Penn group's LLM-agent language with automatic parallelisation, conformal-prediction reliability bounds, and approval-gated security; LLMs write a Python subset that transpiles to Quasar.
N/A (paper-only) · N/A (paper-only)
Infrastructure that operates around these languages, or candidates that haven’t shipped enough to classify.
A typed language for the wiring between agents. Symmetric monoidal category, typed channels, structural morphisms, agents as stateful morphisms with control ports.
Not publicly disclosed · Native binaries (Linux x86_64, macOS Apple Silicon)
Zig-superset systems language with event continuations, mandatory branch handling, phantom typing, and purity tracking. Pre-alpha — 'only ever been compiled on a single computer.' AI-First framing intentionally tongue-in-cheek.
Koru (metacircular, bootstrapped through Zig) · Zig (then native via Zig's backends)
v0.2 design proposal for a language-agnostic IR for agent-driven development. Six specialised agents (Product, Architect, Scrum, Developer, Tester, DevOps) collaborate on shared .spec.ir artefacts; language-specific code generation is downstream.
TypeScript (React POC) · Not applicable — IR artefacts (.spec.ir files), not executable
An AI-native systems programming language announced on Hacker News in March 2026. The Rust MVP compiler ships JSON diagnostics, a JSON AST exporter, a formatter, and a C backend. Public information beyond the repository README is limited.
Rust · C (via the emit-c command)
A community-edited catalogue of programming languages designed for AI agents to author code — not languages that use LLMs at runtime, but languages whose syntax, semantics, or runtime are deliberately shaped around an agent being the primary author.
The bar is intent. A language is in scope if its designers explicitly target LLMs or agents as authors — through token-friendly syntax, mechanically checkable contracts, agent-coordination primitives, or first-class effect declarations for model calls.
A tool that calls an LLM at runtime (chatbots, autocomplete plug-ins, IDE assistants) is not in scope. A language whose only innovation is “works with Copilot” is not in scope. The line is the design intent, not the tooling.
Open a pull request adding a Markdown file to src/content/languages/. The contribution guide asks for evidence the project meets the inclusion bar and a self-classified camp with justification.
The maintainer reviews each submission for fit, accuracy, and tone (the catalogue is descriptive, not promotional). Marginal cases get discussed in the PR thread. Edits to existing entries are welcome — especially corrections from the language's authors.
The catalogue is itself a machine-readable specification. Every detail page has a markdown companion at the same path with a .md extension, discoverable through <link rel="alternate"> and the llmstxt.org conventions.
Use /llms.txt for the short index, /llms-full.txt for every entry in one file, or /sitemap.xml for the full URL set. Agents and crawlers are explicitly welcome.