More and more software companies change to ai vibe coding, and if you tell them, that it is only cut & paste from github and stackoverflow and there for a copyright issue for commercial development, they tell you: "No, AI agents are thinking, they understand, and they create new code" - but do they really?

A study based on esoteric languages proves: No, AI agents are only good with languages where only tons of github and stackoverflow code is available.

The idea

What if you take all the Turing complete programming languages that have the least github and stackoverflow coverage and ask the state-of-the-art LLM models to code different tasks in these languages?

More details on the study can be found here.

The method

They introduced three classes of tasks: Easy like adding two integers together, Medium and Hard, then take the languages with the least repos on github and compare the results to the most popular languages.

They introduced the Esolang-Bench, where they compared the results in the esoteric languages Brainfuck, Befunge-98, Whitespace, Unlambda and Shakespeare and compared it to Python and JavaScript

The models evaluated where: GPT-5.4 xhigh (OpenAI, 2026), O4-mini-high (OpenAI, 2025, OpenAI reasoning model), Gemini 3.1 Pro (Gemini Team, Google DeepMind, 2026, Google), Qwen3-235B (Qwen Team, 2025, Alibaba), and Kimi K2.5 (Kimi Team, 2026, Moonshot).

The result

Turns out, while the ai coding agents are able to solve for example 100% of the easy tasks for the popular languages, even the best model was only able to solve 11% of the easy tasks in an esoteric programming language. No Model was able to solve any task of the medium and hard category.

This proves, that ai is NOT thinking and/or understanding, it is pattern matching existing human code, which means: Using vibe coding for commercial development is a copyright infringement and should bring any commercial software developer using vibe coding in front of a court, paying the original copyright holders.