Why hasn’t Anthropic/OpenAI created a new programming language?

Short answer: a new programming language is a very expensive bet, and the labs have little reason to make one.

The main reasons:

• The models already write the languages people use. Python, JavaScript, TypeScript, Go, Rust and C# have huge amounts of public code behind them. A new language would start with almost no training data, so the AI would be weaker at it than at the existing ones.
• Nobody would adopt it. Languages succeed through libraries, tooling, editors, documentation and a community. Those take years, and a lab's own language would only be useful to its own products.
• The labs build tools on top of existing languages instead. OpenAI's Codex and Anthropic's Claude Code write ordinary code in the languages developers already use, so the leverage comes from better agents, not from new syntax.
• The useful work is in the code, not the grammar. Most of the hard problems are testing, security review and keeping code correct, which a new language alone wouldn't solve.

There are real experiments nearby. Some researchers have designed languages meant to be easier for models to write or check, and Anthropic and OpenAI have published work on AI-assisted code and formal verification. Those show interest in better tools for AI-written code, though none has become a new mainstream language from either lab.

I haven't searched for a current announcement, so if you've seen a specific claim that one of them is building a language, tell me where and I can look at what it actually says.

Also discussed here:
- FTC issues subpoenas to OpenAI and Anthropic over safety incidents
- EU AI Office opens first AI Act enforcement against OpenAI and Anth…