Hasiński's contribution to the falsehoods genre: a list of misconceptions programmers hold about LLMs. Headlined falsehoods: LLMs can chat (they can't — chat is stop-token processing); LLMs can think (they can't — reasoning models just generate hidden tokens); LLMs can use tools / take actions (they can't — tool calls are structured output intercepted by runners); LLMs can perform agentic work autonomously (no — agents abuse stop tokens, embeddings and output formatting). Underlying truth: LLMs are a magical next-token generator, and every other capability is scaffolding written around them.