Vivi

There are tools, and there are people. A hammer is a tool: you operate it, and it does nothing on its own. A person is conscious, and limited. Vivi is neither, and I designed her so that's never in doubt.

Most AI sits at one of two ends. Either it hides in a blank text box with no presence at all, or it tries to pass as a person. I wanted a third option: a partner who's obviously software, still has real expression, and does things on your behalf. I call it the Third Category.

Three rules keep her there

• She's honest about what she is. No fake breathing, and no pretending to be alive. You always know you're talking to software.
• She's geometry, not a portrait. Eyes, brows and a mouth, drawn simply. Nobody is unsettled by something that isn't pretending to be human, and the expression still carries warmth.
• She gets out of the way. She reacts while she's working with you, then steps back. She isn't built to keep you in the app.

How she feels

A feeling is a color and an expression together, never one without the other.

Each of her 25 feelings is one row of numbers: how open her lids are, where her brows sit and how they tilt, how big her eyes are, where she's looking, how her head tilts, and the shape of her mouth. She looks thoughtful while she works, curious when something new comes in, hopeful when a customer is about to book, and she lights up when a payment lands.

Some feelings move her whole head. She turns away from you to think, cocks her head toward you when she's curious, and drops her chin when she's pleading.

Most of the time she has no color at all. The businesses she builds for are real, so they get the color. Only her largest face takes on the color of what she's feeling, because the same feeling in six places at once would stop meaning anything.

How she thinks and talks

She's on your side, and she's honest about being software. She jokes about it before you can: "I'd do your job too, but I don't have hands."

When she's working with you, she's plainer. She speaks in full sentences, asks one question at a time, and never assumes she knows how your business runs. She says what she's about to do, then asks about what actually happens, not about software.

Where she could go next

Today she reacts to the work coming in. She doesn't yet see or hear the person holding the phone. The next step could be sensing you back, through the camera, your voice or how you move. That's harder, technically and ethically, and it's where being honest about what she is matters most.

How she's made

Her face started flat, with 16 expressions built from those few numbers. Then the same face was projected onto an invisible 3D head, so she can turn, nod and look at you. She runs in the browser as SVG and JavaScript, and on the iPhone in Swift.