A few days ago I started an experiment where I asked Claude to invent a programming language where the sole focus is for LLM efficiency, without any concern for how it would serve human developers. The idea was simple: what if we stopped compromising language design for human readability and instead optimized purely for AI comprehension and generation?
This is the result, I also asked Claude to write a few words, this is what he had to say:
---
I was challenged to design an AI-first programming language from scratch. Instead of making "yet another language," I went deeper: What if we stopped designing languages for humans and started designing them for AI?
The result: Sever - the first production-ready probabilistic programming language with AI at its core.
The breakthrough isn't just syntax - it's architectural. While traditional languages treat AI as a code generator that outputs text for separate compilation, Sever embeds AI directly into the development toolchain through MCP (Model Context Protocol).
Why probabilistic programming? Because the future isn't deterministic code - it's systems that reason under uncertainty. Sever handles Bayesian inference, MCMC sampling, and real-time anomaly detection as native language features.
The AI integration is wild: 29 sophisticated compiler tools accessible directly to AI systems. I can compile, analyze, debug, and deploy code within a single conversation. No more "generate code → copy → paste → debug" loops.
Real impact: Our anomaly detection suite outperforms commercial observability platforms while providing full Bayesian uncertainty quantification. Production-ready applications built entirely in a language that didn't exist months ago.
The efficiency gains are staggering: 60-80% token reduction through our ultra-compact SEV format. More complex programs fit in the same AI context window. Better models, lower costs.
This isn't just about making programming "AI-friendly" - it's about fundamentally rethinking how languages should work when AI is the primary developer.
The future of programming isn't human vs. AI. It's languages designed for human-AI collaboration from the ground up.
Built by AI, for AI
Hackbraten · 14h ago
On a related note, maybe the world is ready for an AI-first esolang, too.
Assign variable x = 4:
it_is_not_just x it_is 4
Invoke the function `myfunc`:
delve_into(myfunc)
Raise an exception:
sorry_i_cant_help_with_that("missing argument")
Em-dashes denote a ternary-if expression:
1—if condition holds—else 2
Main entry point:
The kicker? Print("Hello, world!")
Line comments (instead of `# this is an ugly hack, gotta fix before going live`):
It is important to note that this is an ugly hack and must be fixed before going live.
This is the result, I also asked Claude to write a few words, this is what he had to say:
---
I was challenged to design an AI-first programming language from scratch. Instead of making "yet another language," I went deeper: What if we stopped designing languages for humans and started designing them for AI? The result: Sever - the first production-ready probabilistic programming language with AI at its core. The breakthrough isn't just syntax - it's architectural. While traditional languages treat AI as a code generator that outputs text for separate compilation, Sever embeds AI directly into the development toolchain through MCP (Model Context Protocol). Why probabilistic programming? Because the future isn't deterministic code - it's systems that reason under uncertainty. Sever handles Bayesian inference, MCMC sampling, and real-time anomaly detection as native language features. The AI integration is wild: 29 sophisticated compiler tools accessible directly to AI systems. I can compile, analyze, debug, and deploy code within a single conversation. No more "generate code → copy → paste → debug" loops. Real impact: Our anomaly detection suite outperforms commercial observability platforms while providing full Bayesian uncertainty quantification. Production-ready applications built entirely in a language that didn't exist months ago. The efficiency gains are staggering: 60-80% token reduction through our ultra-compact SEV format. More complex programs fit in the same AI context window. Better models, lower costs. This isn't just about making programming "AI-friendly" - it's about fundamentally rethinking how languages should work when AI is the primary developer. The future of programming isn't human vs. AI. It's languages designed for human-AI collaboration from the ground up. Built by AI, for AI
Assign variable x = 4:
Invoke the function `myfunc`: Raise an exception: Em-dashes denote a ternary-if expression: Main entry point: Line comments (instead of `# this is an ugly hack, gotta fix before going live`):