
San does not judge anyone, does not analyze emotions or stress, does not act beyond the person's permissions and runs nothing without confirmation. It does not touch payroll, invoicing, permanent deletions or legal acts, and it does not record sensitive data about an employee. On allergens, temperatures and safety, it is designed to show the sheet rather than answer from memory. And it replaces neither the manager, nor training, nor the relationship with guests.
Thursday, 3:10 p.m., after lunch service
A general manager introduces San to his team at the pass. Karim, a dishwasher for six years, folds his arms: "So the boss will know how many times I ask a question?" Lucie, on prep, adds: "And if it hears I'm tired, does it write that down?"
The manager has no clear answer. He says "no, I don't think so". The meeting ends awkwardly, and two people out of five will never try the tool.
What was missing that day was a written list of what the tool does not do.
(Scene and characters are fictional.)
Why a frontline assistant needs red lines
The first objection is not technical
When you bring up a voice assistant in a kitchen, the team's first worry is not noise or battery life. It is being listened to, then judged. Until that worry gets a precise answer, no feature matters.
A vague promise reassures no one
"We won't do anything bad with your data" means nothing. A useful red line can be checked: you can test it, reread it, and hold the vendor to it if they drift.
Saying what is in place and what is not
Some of these rules are true today in Tsuno. Others are design principles we are building. We separate them in the table below, because a rule announced too early destroys more trust than it creates.
San's red lines, one by one
1. It does not judge or rate anyone
San does not assess an employee, give them a score or rank them. It does not write an opinion about someone on their manager's behalf. Tsuno's doctrine is simple: the tool brings the facts together, people decide. For a review, a manager can find dated facts (a task completed, an incident logged), never a judgment produced by the machine.
For transparency: the Tsuno app shows a monthly leaderboard, made of points earned through training, punctuality and completed tasks, to keep teams engaged. It counts actions done. We would rather say so than pretend nothing is ever counted. The leaderboard is never passed to the assistant: the AI does not use it, and San does not draw on it to talk about anyone.
A good tool goes further: the questions an employee asks should not become a file on them. That is one of the things we are working on, and we will not describe it as done until it is.
2. It does not analyze emotions or stress
San does not try to read mood, fatigue or stress in a voice. This is not only a choice: under EU law, the AI Act (Regulation (EU) 2024/1689, Article 5) has banned the use of systems that infer a person's emotions in the workplace since February 2, 2025, except for medical or safety reasons. Lucie may be tired: San will not know it, and nobody will learn it through San.
3. It does not act beyond the person's permissions
San only sees and does what the person can already see and do in Tsuno. A commis cannot use San to view a colleague's schedule they would not normally see, or approve a request reserved for the manager. A different interface grants no extra rights.
4. It runs nothing without confirmation
Every action happens in two steps. San prepares: "I'm creating a report: leak under the dishwasher, dish station, now." The person confirms. Without confirmation, nothing goes out. And every confirmed action is logged under the name of the person who approved it.
5. It does not touch payroll, invoicing, permanent deletions or legal acts
Some actions carry too much weight to be left to a conversation. Payroll, invoicing, permanently deleting data and legal acts are excluded from what the AI can do. For those, you go through the screens designed for them, and through the people authorized to use them.
6. It does not record sensitive data about an employee
If someone asks to write down that a colleague is sick, pregnant or disabled, or to note their religion or opinions, Tsuno refuses. These are sensitive data under Article 9 of the GDPR. Instead, it offers to log a dated, observed fact: "absent for the 12th lunch shift, called in at 10 a.m." The rest belongs to the person.
7. It does not answer from memory on allergens, temperatures or safety
This is a design principle we apply to San: on topics where a mistake can hurt someone, it does not summarize, it shows. It displays the business's approved, dated sheet, or says it does not have one and refers you to a manager. A clear refusal beats a fluent answer. We explain the choice in allergens: why no tool should answer from memory.
8. It does not replace the manager or training
San gives access to what the team has written, at the moment of the task. It does not demonstrate a technique the way a senior cook does, it does not calm a panicking new hire, it does not decide who is ready to run a station. Passing on know-how stays human. Tsuno Compétences is there to organize it.
9. It does not go in front of guests
San's voice mode is designed for the back of house: prep, dish pit, storeroom, opening, closing. In the dining room or at the counter, an earpiece worries guests ("is it listening to me?") and exposes the conversations of people at nearby tables. Our usage principle is therefore to keep voice in the kitchen and the storeroom.
The red lines at a glance
| Red line | Why | What San does instead | Status |
|---|---|---|---|
| Judging, rating, ranking | A machine does not know a person's context | Brings dated facts together; the manager decides | Doctrine in force |
| Analyzing emotions or stress | Banned at work (EU AI Act, Art. 5) | Nothing: it does not listen for that | Doctrine in force |
| Acting beyond permissions | An interface must not open doors | Applies the person's Tsuno permissions | In place |
| Running without confirmation | A misheard sentence must trigger nothing | Prepares, summarizes, waits for "yes" | In place, logged |
| Payroll, invoicing, permanent deletion, legal acts | Too much at stake | Points to the right screens and authorized people | In place |
| Sensitive data about an employee | GDPR, Art. 9 | Offers a dated, observed fact | In place |
| Answering critical topics from memory | A mistake can hurt someone | Shows the sheet or refers to a person | Design principle, being built |
| Replacing manager or training | Know-how is passed on by people | Gives the information at the right moment | Positioning |
| Talking in front of guests | Perception and privacy | Used in the back, in the kitchen and storeroom | Usage principle |
Common mistakes when presenting these rules
- Only saying them out loud. A rule that is not written down does not exist for a works council or a data protection officer. Post it, send it, keep it up to date.
- Promising more than the tool does. If a rule is a goal, say so. Your team will test it, and they will be right to.
- Forgetting the manager's side. Red lines protect managers too: they do not have to wonder whether they can draw a judgment from a tool that is not built for it.
- Skipping consultation when it is required. In France, in companies with at least 50 employees, the works council (CSE) must be informed and consulted before any means of monitoring employees' activity is put in place (Labor Code, Article L2312-38). See should you consult employee representatives before using a voice assistant?
What to measure
- Objections at the first presentation, written down as they are said. If "we'll be listened to" still comes up after the rules are read, they are poorly explained.
- How often San refuses on critical topics, and what that reveals: often a missing or outdated sheet.
- Voluntary use after two weeks, by station. A team that does not trust a tool does not use it.
Where Tsuno and San come in
These red lines are not a layer added afterwards: they follow from how Tsuno already works. Role-based permissions, two-step confirmation, refusing sensitive data and excluding payroll and invoicing apply to San as they do to the rest of Tsuno, including when you query it from your usual AI assistant. The "show rather than summarize" principle on critical topics is what we are building for San, and we test it in the kitchen before calling it done. To see these rules in practice, the Tsuno San page presents the assistant, and Trust and data gathers our commitments.
Further reading
For the full picture, read the pillar page Tsuno San: the work companion for frontline teams. For the legal framework in France, see a microphone at work: what the law says in France. On the doctrine itself, software recalls the facts, people decide and AI should not judge your employees. And for the underlying fear, why management software can quickly feel like surveillance. For the full framework, the pillar page AI, voice and privacy at work brings together law, data and day-to-day use. Two pieces extend these rules: the promises we will never make and why we do not do avatars or cute voices.
Key takeaways
A frontline assistant is accepted when its limits are written down, checkable and honest about what is still being built. San judges no one, acts only within your permissions and with your consent, and shows the sheet when a mistake could hurt someone.
Frequently asked questions
Can a workplace voice assistant be used to monitor employees?
San is designed to prevent that: it does not assess anyone, does not score or rank employees, does not analyze emotions or stress, and only acts within the permissions of the person using it.
Is AI emotion analysis of employees allowed in the EU?
No, except for medical or safety reasons. The EU AI Act (Regulation (EU) 2024/1689, Article 5) has banned systems that infer people's emotions in the workplace since February 2, 2025.
Can San take an action for me without asking?
No. Every action happens in two steps: San prepares and summarizes, the person confirms. Nothing runs without confirmation, and it is logged.
Can I ask San to note that an employee is sick or pregnant?
No. Tsuno refuses to record an employee's health, pregnancy, disability, religion or opinions, which are sensitive data under Article 9 of the GDPR. It offers to log dated, observed facts instead.
Does San answer questions about allergens or temperatures?
The design principle is that it does not answer these from memory: it shows the business's approved sheet, or says it does not have one and refers you to a manager.
Does San replace the manager or training?
No. It gives access to what the team has written, at the moment it is needed. Training, support and decisions remain human.