Smoke an LLM

Mon, 24 Aug 2026 21:07:04 +0200

I try to keep a tab on what people are doing with Large Language Models (LLMs). It has become more difficult to understand what’s going on. There many different groups and scenes in realm of “ai”. You have vibe coders, agentic engineers, luddites, token maxxers and so on. The more I read about these groups, the clearer it became. This is about smoking. Here me out.

As far as I know there are only two industries that call their customers “users”: IT and drug dealing. In the internet we are still “users”, never humans, persons or people. In the age of “ai” we consume tokens. How many token can you through? The language in that regard is clear.

ai is to internet what tobacco is to our lungs.

If we are still users of the ai tobacco, then what type of users are we? I identified these types of smokers:

Addicted

Uses Claude or another proprietary stack. All-in on token maxxing. Every problem is an nail and Claude is the hammer. Fears the usage limit.

Passive Smoker

Acts as a meat proxy to the chatbot interface. Likes to use Gemini and ChatGPT. Rather asks a bot than a human.

Paper Roller

Uses the OpenCode, Hermes or pi. Heard about OpenRouter. Works at the same speeds as before.

Clean

Has never smoked “ai”. Not even once. Has a “human made” banner in the footer of his indie blog.

Vaper

Runs his own models on an Apple Mini. Thinks this the future. Claims that he is not a smoker.