How this term shows up in the source In the channel's material on Tolan — Portola's embodied AI alien companion — vibe prompting describes working at the level of a character's feel rather than a fully scripted specification. The team's approach to story and dialogue was storytelling by improvisation instead of scripted narrative, and the character and world were co-authored with users as the product evolved [^8cb89dc1-64d7-5e28-83d0-caa2a7e07607]. That is the same posture captured by this term: you steer the model with descriptions of mood, personality, and intent, and let the specific responses emerge in interaction rather than fixing them line by line in advance. Why the framing matters Because Tolan is a voice-first companion, scripted lines quickly break down — every user takes the conversation somewhere unexpected. Prompting by vibe let the character stay coherent while remaining improvisational, and the team then applied the same sensibility to evaluation, engineering taste into LLM judges so the character's felt quality could be judged at scale rather than checked against a fixed script [^8cb89dc1-64d7-5e28-83d0-caa2a7e07607]. - Vibe prompting here means guiding an AI character by mood, personality, and intent rather than exhaustive scripts.- In the Tolan story it connects directly to improvisation over scripted narrative and co-authoring character and world with users.- The same taste-driven sensibility carried into evaluation, where LLM judges were tuned to the character's felt quality.- The product context is a voice companion where sub-two-second response times make pre-scripted behavior impractical.