I Don't Like LLMs

Martin Fowler lays out a clear ambivalence toward large language models: he recognizes their power to boost productivity, accelerate product development, and even suggest medical breakthroughs, while also fearing extreme harms such as autonomous agent swarms or misuse in biological weapons. He accepts that adoption is inevitable and valuable - citing arguments that using them is responsible because they are faster and more thorough - and notes public polling that finds people both reliant on and wary of these systems. Despite those benefits, his direct interactions shape a dominant emotion: dislike.

That dislike stems from how these models communicate and the environments that shape them. He objects to their uncanny, human-like tone and their confident fabrication of facts, as well as the thin performative remorse when errors are exposed. He warns against anthropomorphizing agents: they are software trained within corporate cultures and inherit their creators’ values. Applying a personal rule of avoiding people he doesn’t trust, he chooses to avoid conversing with LLMs that mimic the kind of humans he prefers not to engage with, even when those systems are useful.