Do We Need a New Programming Language for the AI Era?
2026-02-05
AI can now port code between programming languages with surprising accuracy. This capability raises an intriguing question: should we create new languages specifically designed for directing AI?
The Assembly Code of Tomorrow
Alan Kay once observed that “In a ‘real’ Computer Science, the best languages of an era should serve as ‘assembly code’ for the next generation of expression” (timestamp 31:50 in his talk). We may be approaching exactly such an inflection point.
The English Problem
As an engineer (P.Eng, PEO) and programmer, I find it frustrating to specify code using natural language. English lacks precision. When I ask an AI to generate code, I’m forced into the awkward position of describing what I want in a medium designed for human conversation, not technical specification.
What if instead we could write instructions in a language more precise than English, yet flexible enough that AI could translate it into any target programming language we need?
Introducing .rt: A Meta-Language Experiment
I’ve been experimenting with exactly this idea using a meta-language I call .rt, built with a tool called t2t (based on OhmJS and a rewriting-rule DSL I call .rwr).
The concept is straightforward: write once in a clear, specification-focused syntax, then let AI—or automated tooling—transpile it into whatever target language you need.
So far, I’ve written an entire PBP (Parts Based Programming) kernel in this meta-language—about 1,000 lines of code—which generates working implementations in Python, JavaScript, and Common Lisp.
How .rt Works
The syntax resembles Python but uses brace brackets instead of indentation. When operations differ significantly across target languages, I use “macros” as escape hatches. For example:
#read_and_convert_lnet_file (pathname, filename)
This macro represents a read-and-convert-to-json operation that compiles to very different code sequences in each target language.
Currently, I build the .rt meta-language manually. The source code lives on GitHub in four files: 0d.rt, stock.rt, external.rt, and kernel_external.rt, in the kernel subdirectory of the pbp-dev repository.
The AI Question
Here’s what I’m wondering: Can we write a sufficiently detailed prompt that would allow AI to correctly transpile .rt code into various target languages—treating those target languages as mere “assemblers” for higher-level specifications?
If AI can already port code between existing languages, perhaps it could also learn to translate from a purpose-built meta-language designed specifically for clarity and AI-readability.
What This Could Mean
This approach could give us the best of both worlds:
Precision: Write specifications in a syntax designed for exactness, not conversation
Flexibility: Generate implementations in any language the AI understands
Clarity: Focus on what the code should do, not the idiosyncrasies of how each language expresses it
We might be standing at the threshold of a new era where programming languages serve less as direct implementations and more as compilation targets—true assembly code for AI-assisted development.
The question isn’t whether this is technically possible. AI has already demonstrated it can perform these translations. The question is whether we can design meta-languages that make this process reliable, predictable, and practical.
What do you think? Is this the direction programming should take, or are we better off teaching AI to understand our existing languages more deeply?
The .rt meta-language experiment is ongoing. If you’re interested in exploring or contributing, check out the pbp-dev repository.
See Also
Email: ptcomputingsimplicity@gmail.com
Substack: paultarvydas.substack.com
Videos: https://www.youtube.com/@programmingsimplicity2980
Discord: https://discord.gg/65YZUh6Jpq
Leanpub: [WIP] https://leanpub.com/u/paul-tarvydas
Twitter: @paul_tarvydas
Bluesky: @paultarvydas.bsky.social
Mastodon: @paultarvydas
(earlier) Blog: guitarvydas.github.io
References: https://guitarvydas.github.io/2024/01/06/References.html
I share the same feeling about AI programming. A programming language is a middle ground between English and the language of machines. And now we add a layer of NLP to collect ideas from users and turn them into code. Most users lack expressiveness. It's weird. When I was a PM it would take me three months to produce a decent RFP 😁 I don't know if we need a new programming language. I feel that we need a language for specifications, an old idea, I know. I will check your new meta-language. It will take some time to grasp the idea.