
A work assistant needs no avatar, no sense of humour and no seductive voice. People ask it for a procedure, a task or a routine check in the middle of service: it should answer briefly, in a neutral tone, say where the information comes from and admit when it does not know. A likeable character creates something else: trust placed in the tone rather than the source, sometimes attachment, and the illusion that someone is there. That is a design choice we stand by for Tsuno San, and this article explains why, with the texts to back it up.
Friday, 6:05 pm, the dining room before opening
The chairs are still upside down on the tables. Armelle, a server, is setting up her section with the earpiece in. She has just asked the assistant what is left on the front-of-house opening checklist. Then, for a laugh, because the silence is long and Bertrand, the manager, is counting covers at the bar, she says: "And you, how are you doing?"
Bertrand looks up. He can picture two answers. The first sounds like a colleague: "Great, thanks Armelle! You ready for tonight?" It gets a smile, once. The second sounds like a tool: "I'm an assistant, I don't have moods. You have two items left on the opening checklist." It gets a smile too, of a different kind, and Armelle goes back to her tables.
The second answer is the one we aim for. It offends nobody, pretends nothing, and brings her back to service in one sentence.
Why so many assistants have a personality
Consumer products sell company
An entertainment assistant is built to keep you coming back. A first name, a warm voice, a few jokes, a habit of asking "and you?": all of it stretches the conversation. For a product measured on time spent, that makes sense.
Work asks for the opposite
In a kitchen or on the floor, every second spent talking to the tool is a second not spent on a guest. The ideal assistant is the one you forget between two questions. A voice that keeps the chat going, flatters or tries to be witty costs time and attention, both scarce at 8 pm on a Friday.
A friendly voice makes an answer sound more reliable than it is
An answer delivered warmly and confidently inspires trust, whether it is right or wrong. At work, trust should rest on the cited source: the product sheet, the validated procedure, the dated log. If the tone does the job of the evidence, the team ends up believing the tool because it is nice. That is exactly what we want to avoid, especially on allergens and temperatures.
The more it resembles someone, the more human traits we lend it
The question is not new. In November 2021, France's national pilot committee for digital ethics (CNPEN) published its Opinion No. 3, "Conversational agents: ethical issues", which warns in particular about our tendency to attribute human traits to these machines. We draw one simple conclusion for a work tool: it has no reason to encourage that projection.
What EU law says
Article 50 of the EU AI Act (Regulation (EU) 2024/1689) sets a transparency obligation. The provider of an AI system intended to interact directly with people must design it so that they are informed they are interacting with an AI, unless that is obvious to a reasonably well-informed, observant and circumspect person, given the circumstances and the context. The information must be clear and given at the latest at the first interaction. This obligation has applied since 2 August 2026; the full timeline is in AI at work: what changes for employers in 2027.
An avatar that imitates a colleague does not breach this text in itself. But the more an assistant resembles someone, the less obvious it becomes that it is a machine, and the more effort it takes to remind people it is only a tool. We would rather it were obvious from the start.
The same regulation has banned, since 2 February 2025, systems that infer a person's emotions in the workplace, except for medical or safety reasons (Article 5). An "empathetic" assistant that adjusted its tone to the tiredness it detected in Armelle's voice would be on the wrong side of that line. San does not analyse emotions; the details are in AI and emotions at work.
The rules we hold ourselves to
- No face, no animated avatar, no character. San is a product name, and the interface shows what it says and can show where it got it from.
- A clear, neutral voice, chosen to be understood over kitchen noise, not to charm.
- Short answers: one or two sentences out loud, the detail on screen or on the sheet.
- No simulated emotion. No "I'm delighted", no "happy to help, buddy", no "and you?" to keep things going.
- No house humour. If someone jokes, the assistant stays polite and gets back to work.
- An honest answer to "who are you?": an assistant connected to the site's information in Tsuno, acting within the person's permissions.
- Source before tone: the sheet, procedure or dated fact behind the answer, and "I don't know" when nothing backs it.
These rules describe what we want and what a good work tool should do. They do not rule out manners: a neutral assistant can still say hello and thank you.
Work voice versus entertainment voice
| Criterion | Entertainment voice | Work voice |
|---|---|---|
| Goal | Keep you there, bring you back | Get you back to your station fast |
| Length | A conversation that stretches | One or two sentences |
| Tone | Warm, chummy, upbeat | Clear, neutral, polite |
| Humour | Sought after | Absent, without coldness |
| Emotions | Simulated to build a bond | None, neither simulated nor analysed |
| Identity | Character, first name, sometimes a face | A named tool that says it is an assistant |
| Trust | Carried by the voice | Carried by the cited source |
| When it doesn't know | Sidesteps gracefully | Says so, points to the sheet or the person in charge |
| In noise | Rich voice, effects | Crisp diction, steady pace |
Common mistakes
- Confusing neutral with cold. A neutral voice can be pleasant; what you remove is the charm offensive, not the manners.
- Banning the team from joking with the tool. Armelle is allowed to laugh. What matters is that the tool does not play along so far that people start believing someone is there.
- Choosing an assistant because it is likeable in the demo. The demo lasts twenty minutes; service lasts years. Judge the sources and the brevity, not the charm.
- Letting the tool tailor its tone to each person. Adapting vocabulary for a beginner is useful; guessing their mood is not, and may be banned.
- Introducing the assistant as "the newest member of the team". The line is tempting at launch. It plants exactly the illusion you want to avoid.
What to measure
- The average length of a spoken answer, over ten real questions from one service.
- The share of answers that state their source or point to a sheet.
- What the team says, in an anonymous questionnaire, to "do you sometimes feel you are talking to someone?".
Where Tsuno and San fit in
Tsuno San is the intelligent voice assistant for frontline teams, connected to your site's procedures, tasks, checklists and routine checks. On the hardware side, a small dedicated Android device and a light earpiece are all it takes. San stays within the person's rights, acts only once they confirm, and evaluates no one (see what San never does). San Voice is available in demo; San Audio and San 360 are under exploration and testing. The Tsuno San page covers the whole picture, and the trust page explains where the facts behind each answer come from.
Further reading
The same candour applies to our misses: read why we show our failed trials. For the difference between a consumer assistant and a work tool, see general voice assistant or a companion that knows your workplace. And for where the tool stands next to the person doing the work, a companion, not a supervisor. For an overview of the tool, read the pillar page Tsuno San, the work companion for frontline teams.
Key takeaways
At work, an assistant earns trust through its sources and its brevity, not its voice or its jokes. When Armelle asks how it is doing, the right answer reminds her it is a tool and gets her back to service in one sentence.
Frequently asked questions
Should a workplace voice assistant have a personality?
No. At work you ask it for a procedure, a task or a routine check. It should answer briefly, in a neutral tone, say where the information comes from and admit when it does not know. A character, jokes or a seductive voice add nothing and can create attachment or misplaced trust.
Do employees have to be told they are talking to an AI?
In the European Union, Article 50 of the AI Act has applied since 2 August 2026. It requires the provider of an AI system intended to interact directly with people to design it so that they are informed they are dealing with an AI, unless that is obvious, at the latest at the first interaction. Outside the EU, check local rules.
What should an assistant say when someone asks how it is doing?
One short, honest sentence: it is an assistant, it has no mood, and it offers to get back to work. It should neither fake an emotion nor turn the question around to keep the conversation going.
Why would a warm voice be a problem?
Because a friendly, confident voice lends weight to everything it says, right or wrong. At work, trust should come from the cited source, not the tone. A clear, neutral voice is also easier to understand in a noisy kitchen.
Can the assistant read the mood of the person talking to it?
It should not. In the European Union, the AI Act has banned systems that infer people's emotions in the workplace since 2 February 2025, except for medical or safety reasons. San does not analyse emotions.