Kristian Polso writes why he will not use LLMs in his coding

A developer with ten years of production machine learning work says he will not use LLMs to write his code. Kristian Polso posted the piece on his blog on 8 November 2026. By the end of the first two hours on Hacker News it had reached 81 points and 146 comments.

Polso writes PHP and JavaScript and runs his own business in Finland. He says he does not reject the technology itself. His piece sets out five reasons he opts out of using LLMs in his own work. He gives sources for the strongest claim.

The five reasons he lists

The first is cognitive. Polso cites a Microsoft Research paper on how AI tools cut cognitive effort for knowledge workers. He also links an arXiv paper on cognitive debt from AI essay assistants. He writes that heavy LLM use leaves him feeling "disoriented and confused".

The second is code quality. He says LLM output looks polished and reads clean at first, and often has unneeded complexity under the surface. He points to his own experience reviewing Claude output at work. He cites an article titled "LLMs expose the literacy gap between great and mediocre coders".

The third is maintenance. He writes that if agents write all of the code, no human will understand it over time. He frames this as a slow risk, not a sudden one.

The fourth is personal. "I just simply love writing code. It's part of my process of problem-solving," he writes. He treats this as a preference and not a rule for others.

The fifth is environmental and legal. He cites the public Finnish broadcaster Yle on Google cutting forest for a data centre in Finland without a permit. He also raises the US CLOUD Act. That law gives US agencies access to data held by US companies abroad.

Where the piece sits in the wider debate

Polso frames his position as personal. He writes that he respects engineers who choose differently and that LLMs are a real technology. The Hacker News discussion is split, with the top comments arguing both sides. Few recent pieces from an experienced engineer say no in this much detail, with sources.

A team standardising on a coding agent this quarter has something to measure against here. Polso's cognitive point is testable on your own work. Track where the agent cuts review time and where it adds to it. Make that split visible before the next tool rolls out across more engineers.

Source

Why I will not be using LLMs in my coding by Kristian Polso. Discussion on Hacker News.