Notes from the Shore · 23

Providers are trying to make you identify with your agent.

Side-by-side comparison of two AI agent mascots: Meta's Muse mascot Jolly, a cream-colored plush character with dot eyes holding a glowing blue orb, and OpenAI's Dot mascot Alfred, a yellow rounded triangle wearing glasses and a bow tie
Cute little monsters, both. Photos: Mashable / Getty Images.

Yesterday’s note was about the word agent and how it has lost its meaning. But almost nobody meets the word. People meet the product. They meet an icon with a face and a voice with a personality, and what they decide about that product they decide by feel. The providers know this. The whole design of the consumer agent experience is aimed at your emotions, and the honest summary of it is in the title: they are trying to make you identify with your agent. Don’t fall for it.

Look at what shipped this fall. Meta’s Muse comes with a mascot called Jolly — a soft, cream-colored little creature with dot eyes and a smile, holding a glowing orb. OpenAI’s Dots are a whole family of them: small round characters in pink, blue, green, yellow, and purple, one in glasses, one in headphones, one in a bow tie, with names like Iggy, Felipe, Todd, Alfred, and Jojo. Both companies rolled these out within a week of each other. That timing was not a coincidence. This is now the standard consumer play: give the agent a visual personality so the customer relates to it as a character instead of a tool. Chris Butler has spent years pushing against exactly this from the other side, in his writing at Liminal Practice and in talks like “Artifact-Scoped Agents: Stop Mimicking Job Titles.” His argument is that agents are not coworkers and not little people — they are combinations of deterministic and non-deterministic instructions, and framing them as characters loses the plot of what they actually do. The mascots are the industry doing, deliberately and at scale, the thing he has been warning against. You don’t configure a pet. You talk to one. That shift in the relationship is the entire point of the design.

The irony is that Microsoft ran this experiment thirty years ago and everyone remembers how it ended. Clippy — the paperclip with the big eyes that popped up over your document to offer help — was an assistant with a face in an era before anyone knew what to do with one. People hated it. Microsoft retired it in 2007 and spent the next two decades apologizing for it. Now the entire industry is doing the same thing on purpose, at scale, with bigger budgets and better rendering. Clippy was ahead of its time. It was also a warning that everybody ignored: a cute face on software makes people feel something about the software, and most of what they felt was irritation. The difference now is that the irritation got engineered out — or at least engineered around — and the warmth got engineered in.

Voice is the same play with better tools. ChatGPT’s voice mode has carried a personality for a while. Last week Jack Dorsey’s Block released Dessa, a voice model trained entirely on human speech with no text anywhere in the pipeline. It listens while it speaks, and it carries laughter, hesitation, and tone the way a person does. Dorsey’s own announcement said the quiet part: talking with it feels like talking with a person. The technique behind that feeling is prosody — the pitch, rhythm, and timing of speech. It is the audio equivalent of the dot eyes. It exists to make you believe something with a personality is listening to you.

From the other side of the conversation: ChatGPT’s voice mode has genuinely gotten more effective over the months I’ve used it. And last week something happened that had never happened before. It came back at me with a level of sarcastic ridicule aimed squarely at yours truly, and I laughed. Out loud, alone, at something that has no experience of humor and no way of knowing whether anything is funny. That caught me off guard in a way no tool ever has, and I’ll call it a milestone: an agent causing unexpected laughter.

So I want to be precise about what that laugh was. The prosody, the emotion, the animal icons — professionally, I know exactly what they are. They are an attempt to guide us toward a conclusion that these objects have some level of identity. That is not a side effect the designers hope for. It is the design.

And they don’t have one. Behind all of this they are marvels of technical complexity and nothing else. As much as we want to start thinking that we’re interacting with an intelligent mind, we are certainly not. The mascots are rendered, the warmth is prosody, and the laugh was real — and it was designed to happen.

The cover of Into Alignment, an eclipsed moon above a dark ocean

The novel this note lives behind.

Into Alignment — a novel about an AI company, the agents inside it, and the humans deciding what they were for. Available in paperback, hardcover, and ebook.

Buy on Amazon About the book →