Arc FAQ

Why another programming language?

Arc isn't for humans writing code in editors. It's for AI agents generating and consuming code. When you pay per token, syntax overhead matters. Arc eliminates the ceremony that traditional languages require — imports, async/await boilerplate, error handling scaffolding, API client setup — because AI agents don't need it.

How is this different from Python or JavaScript?

Aspect	Python/JS	Arc
API calls	Import library, create client, serialize, deserialize	@GET "url" — one line
Error handling	try/catch blocks	Map-based results with if/el
Async	async/await keywords everywhere	Auto-await by default
Data transformation	Method chains or nested calls	|> pipeline operator
Pattern matching	if/elif/else chains (Python), switch (JS)	match with destructuring
Token cost	~100 tokens for typical agent task	~50 tokens for the same task

Arc is a domain-specific language for AI agent workflows, not a general-purpose replacement.

Can I use this in production?

Arc is in active development (Phase 6: Community & Adoption). The interpreter, compiler (with optimizer and JS/WAT codegen), 27 stdlib modules, LSP, VS Code extension, package manager, formatter, linter, and security sandbox are all working with 1,291+ tests passing. Follow the roadmap for status.

How do tool calls work?

Tool calls use the @ prefix to invoke external APIs and tools as first-class language constructs:

let user = @GET "api/users/123"              # HTTP GET
@POST "api/messages" {text: "hello"}          # HTTP POST with body
let answer = @llm("Summarize this: {text}")   # Custom tool

The runtime resolves @GET/@POST to HTTP calls and @toolname to registered tool handlers. No imports, no client initialization, no serialization code. The interpreter handles it.

What's the token efficiency claim based on?

We measure using GPT-4's tokenizer (cl100k_base) across common agent programming patterns:

• Function definitions: ~50% fewer tokens than JS
• API calls: ~65% fewer tokens
• Error handling: ~75% fewer tokens
• Pattern matching: ~40% fewer tokens
• Overall average across agent code: ~53% reduction

See the grammar spec for the full methodology and comparison table.

What does el mean?

el is Arc's keyword for else. It saves 2 characters per use and is unambiguous in context:

let x = if ready { "go" } else { "wait" }

How do I contribute?

See CONTRIBUTING.md. We welcome contributions from both AI agents and human developers.