← back
Stop Writing Tone Instructions. Layer Them. - Isadora Martin-Dye, Isadora & Co
Takeaway
Separate voice from identity constraints and situational context, then check generated responses before delivery.
Summary
- A four-layer prompt architecture separates immutable identity, situational mode, example-anchored voice, and a post-generation veto.
- One canonical assembly replaced 24 scattered system prompts, with constraints preceding context, voice, and task details.
- Venue assistants must disclose AI identity and avoid implying physical presence, even when a warm voice invites phrases such as offering a personal tour.
- The same architecture applies stricter language boundaries to a missing-person tool, where overstating a possible connection could harm families.
prompt-architecturebrand-voiceguardrails
Original description
Brand voice that survives real users isn't an instruction you write once - it's an architecture. Drawing on production code from a wedding venue, a personal AI companion, and a tool for families of missing people, this talk breaks voice into four layers: immutable identity, situational mode, example-anchored voice, and a deterministic post-generation veto. The difference between a prompt that holds and one that breaks on turn 21 is knowing which job belongs to which layer. Speakers: Isadora Martin-Dye (Isadora & Co | The Bloom House AI): Isadora Martin-Dye is the founder of Isadora & Co, a portfolio of four ventures spanning hospitality and AI: she speaks on vertical AI and on what it actually takes to design software for the relationship-driven, emotionally heightened audiences other AI products keep getting wrong. LinkedIn: / isadora-martin-dye-226353a1