Continuously Integrating Feelings: processing feelings moment to moment for reflectively stable policy changes

Johannes C. Mayer published a detailed account of using Anthropic's Claude Opus 4.6 (via Claude Code) as an introspective thinking partner to process unexplained negative emotions. The 23-minute conversation, which took place on March 1, 2026, demonstrates a novel application of LLMs beyond typical Q&A—using the model to check written sentences against emotional responses and collaboratively build a psychological model. Mayer moved from feeling "really bad" with no clear cause to developing what he calls "Continuously Integrating Feelings," a framework arguing that heuristic brain algorithms manifest as feelings containing valuable information that analytical thought wouldn't generate.

The process yielded multiple concrete outputs: a formal document outlining why emotional integration increases effectiveness, a custom-generated song designed to be intrinsically rewarding (created so Mayer would engage with the content repeatedly), and an Emacs-based reminder system. The work operates on four levels: object-level emotional modeling, process-level LLM partnership techniques, implementation-level tool building, and meta-level demonstration of the very method being discovered. This represents a significant shift from using AI for external problem-solving to employing it as a catalyst for internal emotional processing and reflective stability in decision-making.