I will not be using LLMs in my coding

After years building production machine-learning systems and experimenting with local language models, Kristian Polso has decided not to use LLMs or AI agents in his professional coding. His objection is not that the technology is unfamiliar or wholly useless: he acknowledges that an AI assistant can sometimes suggest a way forward when he is stuck. But he judges that occasional help insufficient to offset what he sees as significant costs. He cites research on reduced cognitive effort and “cognitive debt,” and says extended use leaves him feeling confused and disoriented.

Polso also argues that fluent, elaborate generated code can conceal poor solutions: he has reviewed AI-written changes that were needlessly complex, overbuilt, and padded with excessive tests, while less experienced programmers may struggle to spot the problems. He worries that agent-written systems will become difficult for humans to understand and maintain, and says writing and reading code are themselves valuable parts of reasoning and understanding people’s decisions. His other concerns include the environmental impact of data centers and the privacy and political risks of relying on US-based companies, especially under the US CLOUD Act despite EU data-storage assurances. He rejects the idea that AI use is inevitable: the tools may persist, but developers can choose not to adopt them. His decision is personal, not a demand that others stop using AI.