Somewhere in most offices there is a plant with a name on a strip of tape. Somebody named it, nobody voted on it, and now three people ask whether anyone watered Kevin this week.
This is one of the most reliable things human beings do, and it has nothing to do with technology. Sailors name ships. Drivers name cars. Musicians name instruments. On a shop floor a machine gets a name within a month of arriving, usually an unflattering one, and after that the operators will tell you it does not like Mondays.
None of these people believe the object is alive. Ask them directly and they will laugh at you. They do it anyway, and it is worth understanding why, because the same move is now happening in front of software that can answer back, and almost nobody is designing for it deliberately.
The finding that makes this hard to dismiss
In 2007 four researchers ran a study on people who owned a robot vacuum cleaner. Thirty owners, average ten months of ownership, interviewed about how the thing had settled into their homes.
Twenty-one of the thirty had given it a name. Sixteen spoke about it as a he, and when asked why, pointed to its shape, its colour, and an assumption that machinery is men's territory; two women said plainly that they liked the idea of a man doing the cleaning. The naming itself most of them could not explain. They said it felt natural, and that the machine had earned one.
A vacuum cleaner. A disc that bumps into furniture and goes back to its dock. There is no face, no voice, no words, nothing that could reasonably be mistaken for a mind.
The researchers' own conclusion is the part worth keeping. They argued the naming was what allowed people to talk and write about the robot at all, and that talking about it was how the relationship formed. The name came first. The communication followed. The relationship was built out of the communication.
That order matters, and most people get it backwards. We assume characterization is the decoration you put on a relationship once it exists. It is closer to the opposite: characterization is the thing that makes communication possible, and communication is what builds the relationship.
What naming actually does to a person
To man the world is twofold, in accordance with his twofold attitude.
Buber opens his book by saying a person can stand toward whatever is in front of them in exactly two ways. In one, the thing is an It: something you handle, classify, use, get a result out of. In the other, the thing is a You: something you address. Same object either way. What changes is the stance, and everything about the encounter changes with it.
Nobody consciously decides to treat an object as an It. That is simply the default whenever the thing gives you nothing to address.
A blinking cursor gives you nothing to address. A dial gives you nothing to address. A machine with a name and a reputation for hating Mondays does, and so people talk to it, and about it, and around it, and they notice when it sounds wrong.
Behavioural research has been saying a version of this for thirty years: people apply social rules to machines automatically, without deciding to, including things as specific as politeness and gender. It is not a mistake people make. It is how attention works. Attributing character is the easiest and most expected result of perceiving something, of making it predictable.
And here is the limit that held for the entire history of this behaviour, until very recently. The ship never answered. You could name it, defend it, blame it, talk to it in a storm, and it gave you nothing back. The whole relationship lived in one person's head. The only payoff was that you treated it better.
What changes when the object can answer?
Language models broke that limit, and the consequence is not sentimental. It is mechanical.
For the first time the thing on the other side responds in language, holds what you told it, and does something with it. Which means the psychological move people were already making for free stopped being decoration and started doing work, because these systems run on precisely the thing that personification produces: more language, more context, more of what you were actually thinking.
Watch two people with identical software. The first types unpaid invoices. The second types "can you pull the invoices still unpaid from last month, I need to know who to call before Friday." The second one gets better work back, and will keep getting better work back for a year. Neither of them knows anything about how the system works. The difference is that one is operating and the other is talking.
Fifty years of software trained the first behaviour into all of us and trained it well. You do not explain yourself to a search box. You learn the shortest input that produces the output and you keep the rest of your reasoning in your head, because the machine has no use for it. That was correct for every piece of software any of us ever used, and it is exactly backwards here.
So the loop closes on itself. A thing you can address gets talked to. Talking produces context. Context produces better output. Better output earns more work. And none of it required the model to get smarter, which is worth saying out loud, since the industry spends nearly all of its attention on the half of that loop that is hardest to move.
What about agents?
An assistant that answers questions is one thing. An agent is something else: it works over time, across channels, with permissions, and it takes actions that have consequences. Naming a vacuum cleaner costs nothing if you are wrong about it. Building a working relationship with something that sends invoices is a different proposition.
It comes down to one line: people delegate to someone, not something. To hand work over you have to be able to predict what comes back, and to predict it you need a consistent identity on the other side.
Which introduces a requirement the psychology does not care about but the architecture has to solve.
If names be not correct, language is not in accordance with the truth of things.
A character is a promise about future behaviour. If the thing behind the name answers differently every week, forgets what it was told, or behaves like a different entity on email than it does on WhatsApp, then the person is addressing something that does not exist. They will feel the mismatch long before they can explain it, and what they do next is the most expensive failure in this business: they stop using it quietly, without ever filing a complaint.
Research on personified systems points the same way. When studies isolate appearance alone and hold behaviour constant, the effect largely disappears. What carries the weight is the verbal and behavioural layer: how it writes, what it consistently asks for, what it refuses, whether it remembers last week, whether Thursday's version behaves like Monday's.
So the face is a handle. Consistency is the thing you are holding. A handle attached to nothing is worse than no handle, because it invites a relationship the system cannot hold up its end of.
How we build for this
This is the problem Botonom is arranged around, and most of the answer is unglamorous plumbing rather than character design.
The identity is built by the customer, not shipped in a box. Eighteen traits before you get to colour, which is why no two companies meet the same agent. A stock character handed to ten thousand companies is a mascot. An identity a team assembled for the work they actually do is something they can address.
One memory across every channel. The same agent on WhatsApp, email, the web and mobile, carrying one conversation and one history. This is the technical requirement that makes the psychological thing hold, and it is where most implementations quietly fail: an assistant that is a stranger again on the next channel cannot be anybody.
Standing rules that persist. Tell it once and it holds, across sessions, indefinitely. That is what consistency means in practice, and consistency is what the name has to be backed by.
Permissions and approval gates. In Botonom the agent's authority is defined at the data layer, and the actions that matter always go through a human. This is the honest counterweight to everything above: someone who treats an agent as a colleague checks it more carelessly than someone who treats it as a tool. The cost of that carelessness is inevitable. The answer is not a blander agent but limits drawn clearly and distinctly, and a transparent, inspectable record of past actions. Botonom was built with this architecture from the start.
The persona describes work, never feelings. This is the rule we hold hardest, and perhaps a slightly dull one. An agent that says it is excited about your project reads as theatre, and people may not be able to name it exactly, but they sense it. An agent that says it does not process refunds over a certain amount without approval reads as a competent colleague. Same budget, spent in two entirely different places.
The point
People were always going to do this. They did it to ships, to lathes, to a vacuum cleaner that bumps into chair legs, to the plant by the window.
What is new is that the object can now answer, and that the quality of what it answers depends on how much the person tells it, and that how much the person tells it depends almost entirely on whether they believe they are operating something or talking to someone.
The person who thinks they are talking builds a working relationship over time. The person who thinks they are operating ends up with a search box that talks, and a quiet conviction that this technology is overrated.
Giving it a name and an appearance is not simple decoration. It is the essential thing that determines which way the relationship will go.
Frequently asked questions
Isn't this just anthropomorphism dressed up?
It is anthropomorphism, and that is the point: it happens whether or not anyone designs for it, as the vacuum cleaner study shows. The design question is not whether to permit it but what it attaches to: a deliberate, consistent identity, or a vague impression a team assembles from fragments over several months.
Does the visual face matter, or is a name enough?
A name does most of the work, since the effect lives in language and behaviour rather than in the picture. The visual identity mainly helps people find and tell agents apart once there are several, and it makes the shift from "the system" to a specific someone happen sooner.
Can people put more trust than they should in an agent that feels like a colleague?
They can, and that is the genuine cost rather than an imaginary one. The counterweight is not a duller agent but clearly stated limits and a record a human can open, so trust rests on something checkable rather than on an impression.
Should the agent express emotion?
Describing what it is doing, what it decided, or what it will not do reads as competence and is safe. Describing what it feels is where personified systems start to feel like theatre. Say what the agent does, never what it feels.
