The author criticizes current AI products like ChatGPT, Gemini, and Claude for not treating their limitations seriously. They argue that these tools present themselves as problem-solving assistants while ignoring their core flaw: hallucinations. The text highlights how error-checking is dismissed as legalese rather than integrated into the workflow. The author proposes a 2-column worksheet interface where users must verify each claim with notes and a checkbox. They also suggest coding assistants should include pre-test diff checks to avoid wasting compute. The piece concludes by demanding more citations and details instead of unsupported answers, calling the current approach a grift.

The author specifically mentions Ollama as another offender due to its lower model quality. They argue that serious tools must provide affordances for verification, not just warnings. The post is inspired by the universal admission that AIs make mistakes, yet no product builds a workflow around this reality.