liszper

Made this for myself, please fork it and build your own harnesses, there is really not a lot to it. The best coding agents are custom-tailored to your needs.

This idea came to me in a dream. The starting point is a neuroscience finding that's been replicated across multiple fMRI studies: programming languages are not languages. When you read code, your brain routes it through the Multiple Demand network, the same circuitry used for logic puzzles and spatial reasoning. The language network stays dark. Your brain sees through the syntactic dressing and recognizes code for what it is: a deterministic puzzle, not communication.

This means no programming language has ever actually been a language in the neurological sense. Even "natural language programming" efforts like Inform 7 fail the test, the programmer still maintains a mental model of state machines and boolean flags, which is pure MD network territory.

SOEL is an experiment: what if the source code were genuine natural language prose — narratives with entities, intent, ambiguity, and social context and the compiler handled the translation to executable code? You write something closer to a specification document than a program. The compiler semantically encodes it, flags genuine ambiguities as compiler errors (which you resolve through dialog, not syntax fixes), and generates GHC-compilable Haskell.

It's impractical, unreliable, and fascinating. The theoretical foundation is in SPEC.md if you want the full neuroscience deep-dive.

hackers can also cook and not become a chef

you can OSINT me pretty easily, not going to post it here for the sake of anonymity against crawlers who train models on our conversations. today's HN comments are tomorrow's coding LLMs

No, It has something to do with experience. The system is highly integrated to other platforms and have to stay afloat during burst loads.

I see myself as a hacker.

200k DAU, 7 million registered, ~50 microservices, large monorepo

I'd argue this just proves my point.

I'm also a lisper, yes.

Yes, exactly. Learning new things is hard. Personally it took me about 200 hours to get started, and since then ~2500 hours to get familiar with the advanced techniques, and now I'm very happy with the results, managing extremely large codebases with LLM in production.

For context before that I had ~15 years of experience coding the traditional way.

I agree with your point. I think this is the reason why most developers still don't get it, because AI coding ultimately requires a "higher level" methodology.

With all due respect, you sound like someone who is just getting familiar with these tools. 100 more hours spent with AI coding and you will be much more productive. Coding with AI is a slightly different skill from coding, similar how managing software engineers is different from writing software.

most SWE folks still have no idea how big the difference is between the coding agents they tried a year ago and declared as useless and chatgpt 5 paired with Codex or Cursor today

thanks for the article, it's a good one

Funky how you can just generate these things in 2 minutes

lmao

Really nice, thanks for sharing!

Agentic AI is when an LLM uses tools. This is a minimal but complete example of that.

Thank you, fixed that!

curl was cheating yes, might go zero dependencies in the future.

Working on minimal local training/inference too. Goal of these experiments is to have something completely independent.

Yes, it is fun to create small but mighty executables. I intentionally kept everything barebones and hardcoded, because I assumed if you are interested in using Agent-C, you will fork it an make it your own, add whatever is important to you.

This is a demonstration that AI agents can be 4KB and fun.

Updated to CC0!