FIELD NOTE 02 / CONVERSATIONAL AI

What building Margo taught us about trust.

A believable voice is useful. A trustworthy conversation is more important.

Margo is Minor Legend's realtime AI receptionist. Anyone can call the studio, explain what they need, ask a question, and experience the system directly. Building that experience made one thing clear: conversational AI is judged less by how impressive it sounds than by how safely and calmly it handles imperfect human behavior.

Speed changes the emotional tone

A delayed response feels artificial even when the words are excellent. Realtime systems must minimize the gap between listening, understanding, and speaking. But speed alone is not enough; the agent also needs permission to pause when the caller is still forming a thought.

Interruption handling is a trust feature

People cough, change direction, speak over one another, and make background noise. A system that treats every sound as a new instruction feels fragile. Thoughtful turn-taking requires balancing responsiveness with enough patience to let the caller finish.

Natural language must still be standard language

AI can generate sentences that are technically correct but socially unusual. Phrases such as “I need context so the recipient will know the context” expose the machinery. The best correction is often not a more elaborate prompt—it is a simpler, more familiar sentence.

Human-sounding is not the same as human. Trust begins with clear disclosure and predictable behavior.

The handoff is part of the conversation

A receptionist is not valuable because it can talk. It is valuable because it moves the call somewhere useful. Margo captures the reason for the call, prepares the next person, and supports transfer, booking, or follow-up without making the caller repeat everything.

Personality should never compete with clarity

A distinctive voice can make the experience memorable, but the personality should remain calm, concise, and situationally aware. The caller's goal matters more than the agent's performance.

Design the failure states first

The system needs graceful behavior when it cannot understand, cannot transfer, loses connectivity, or receives a request outside its role. Good conversational design makes those moments feel controlled rather than mysterious.

Margo is an internal Minor Legend build, not a customer case study. That distinction matters. The purpose of publishing the system is to make our capability testable and to keep improving it through real use rather than presentation alone.