Is there a 'right amount' of AI?

The central question is whether there is a defensible limit to AI-assisted coding. The writer is not opposed to using software that contains any AI-generated code: they use apps whose developer employs coding agents, and accept that a carefully reviewed contribution can be small and harmless. The debate over the game Big Walk makes that position harder to sustain. Its developers said LLMs helped with a few technical problems, touching less than one percent of one percent of the code, yet critics argued that even optional use supports systems built through exploitative practices and displaces human help. The writer sees a meaningful difference between limited, understood assistance and wholesale delegation, but questions why developers who prize control over their code would welcome machines writing it.

The concern extends beyond ethics to skill, reliability, and livelihoods. Coding agents can duplicate functions, add unused code, or invent bugs, while their apparent familiarity with a project is not genuine understanding; greater reliance can make software harder to inspect and maintain. The writer also warns that AI-generated work can be merely adequate, threatening independent developers and technical writers whose work is undervalued when “good enough” passes. Local models and training data with clear permissions might improve matters, but current tools still raise environmental, licensing, and labor concerns. The practical conclusion is to use LLMs deliberately, not assume they are necessary, and recognize that crafted software depends on human judgment and understanding. Zero AI-generated code remains an available option, and the easiest amount to vet is the least amount used.