Built an notebook with an AI assistant that pushes back

A sample brainstorm notes A lot of conversational LLMs suffer from chronic sycophancy. Even when specifically prompted to critique or play devil's advocate, models naturally drift toward agreement, validating weak arguments, and adopting the user's premise within a few turns. Beyond prompting, standard chat UX actively encourages this problem. When an AI generates paragraphs in a conversational back-and-forth, the user naturally shifts into consumption mode rather than critical thinking mode. This in a way becomes a feedback loop where the user becomes dependent on the AI. I built this as a dedicated notebook instead. Rather than a chat interface that dumps walls of text and takes over the writing, you draft your thoughts in blocks while a side assistant helps you guide your logic and actually pushes back. The goal here is to encourage independent thinking with AI assistance Threw together a lightweight, no sign-up prototype if you want to test the workflow:https://paper-dusky-five.vercel.app/