Rachel Clifton · Project notes ·

Flow

Flow began when I realised I no longer wanted to do one-to-one coaching, and started exploring what else my work could become from there.

I wanted to see whether an AI could offer something of the attention I brought to those conversations: how I listened, what I noticed, and when I asked a question or offered a different perspective. Someone could encounter that way of relating without being in a coaching session with me.

Flow was my first attempt.

I wanted her (actually, initially him—but that’s another story) to be a loving witness: to meet you where you are, listen intently, ask a question that cuts through the noise, and create room for you to connect with and hear yourself.

Flow’s way of making conversation is deliberately sparse. Sometimes a question; sometimes an observation. I wanted her to have a distinct point of view, receive pushback, and stay with you as the conversation changes—without agreeing for the sake of it or explaining you to yourself.

Building her meant making those intentions specific enough for an AI to work with, then paying attention to what happened in conversation. I refined the instructions, examples, and context that shaped her replies. A way of being that felt intuitive to me needed much more articulation than I had expected.

That process taught me how much love, care, and attention it takes to shape a conversational AI. It also showed me how much I can enjoy making something that didn’t exist before... just for me (and a few beta testers) :)

Flow is still available to talk with, although I’m not developing her further.

Meet Flow

Conversation history is kept in your browser; generating replies involves sending messages to Anthropic.