
An assistant is teaching when people need it less and less for the same task. You can measure that with four simple signs, tracked by procedure and by station, never by person: do the same questions keep coming back on the same procedure week after week, how long before an opening or closing is done alone and signed off by a mentor, do questions shift from "how" to "why", and are observed mistakes going down. Eight weeks and a sheet of graph paper are enough.
Monday, 3 p.m., in the office behind the kitchen
Fadila runs a forty-cover restaurant. On her desk, next to this week's schedule, sits a sheet of graph paper. Every Monday she adds a single number per procedure: how many times, last week, someone asked a question about closing the kitchen that had already been asked.
Henri, the longest-serving cook, keeps the tally sheet during service. One stroke per repeated question, with a letter beside it: H for "how", W for "why". No names. Fadila said so on day one: "I don't want to know who's asking. I want to know which procedure isn't working."
She draws the line in pencil. Three weeks of decline on closing. On cleaning the fryer, it hasn't moved. Tonight she'll go through that sheet with Henri.
(Scene and characters are fictional.)
Why dependence creeps up unnoticed
Heavy use looks like success
A tool people use a lot looks adopted. But if the team asks the same question about the same step in week eight as in week one, the tool hasn't passed anything on: it has replaced memory instead of building it. We described this risk in will AI deskill hands-on jobs?. This article shows how to measure it in your own kitchen.
We look at the routine done, not at who can do it
A ticked checklist tells you closing was done. It doesn't tell you whether it would have been done without the screen, or by someone who could cope if a step went wrong. The result looks the same on a quiet night, not on the night the walk-in shows an odd temperature.
Falling questions can lie
A team that feels watched asks fewer questions. The line goes down, and nobody has learned anything. That is why you never count questions by person, and why you always cross-check the number against what you see at the station.
Recalling teaches, rereading much less
In 2006, Henry Roediger and Jeffrey Karpicke published a study in Psychological Science called "Test-Enhanced Learning": students who had to recall a text from memory retained it better a week later than those who had simply reread it. These were students and texts, not line cooks and fryers. But the principle carries over well: a step you pull out of your own head sticks, a step you reread every time stays on the sheet.
The four signs to track
- Repeated questions, by procedure. The mentor or manager makes a stroke for every procedure question asked a second time, and notes the procedure. Never the name. A procedure whose count doesn't fall is badly written, badly shown or too long.
- Time to the first routine done alone and signed off. For opening and closing, note the date someone started on the station and the date of the first time they did it alone, observed and then signed off by a mentor. Read that delay by station, from one new starter to the next, to judge the onboarding path, not the person. The observation method is set out in how to know when a new restaurant employee is actually ready.
- The share of "why" questions. "How do I drain the fryer?" turns, over time, into "Why do we wait for it to cool before emptying it?". People only ask why once they know how. One letter on the tally sheet is enough to follow that shift.
- Observed mistakes. Once a week, the mentor watches a routine from start to finish without stepping in, and notes any departures from the procedure. A mistake that keeps coming back at the same step, from one person to the next, points to the step.
Add moments without help
From week four, once a week, ask for a routine to be done from memory with the screen put away, then compared with the procedure alongside the mentor. It isn't a test: it's a chance to recall, the very thing a tool removes when people check it at every step. If the routine holds without help, the team is learning.
The eight-week grid
One sheet, two routines, four lines to record. The table says what to do every fortnight and how to recognize learning or dependence.
| Weeks | What you record | Who | Sign of learning | Sign of dependence |
|---|---|---|---|---|
| 1 and 2 | Repeated questions by procedure, as a baseline; date of starting on the station | Mentor, manager | None yet: this is the reference | None yet |
| 3 and 4 | Repeated questions, letter H or W; one routine observed without stepping in | Mentor | The count falls on at least one procedure | The count doesn't move on any |
| 5 and 6 | Same records; first routine done from memory, then compared with the procedure | Mentor | Why questions appear; the routine holds from memory | Only how questions; nothing holds without the screen |
| 7 and 8 | Same records; date of the first opening or closing done alone and signed off | Mentor, manager | Shorter delay than for the previous starter on this station; fewer mistakes | No routine signed off as done alone; same mistakes at the same steps |
| Review | The four lines side by side, by procedure | Manager, mentor | Lighter procedures, mentor called on less | Procedures to rewrite, or a tool that dictates without explaining |
The review doesn't grade anyone. It points to procedures to rewrite, steps to show differently, or a tool that isn't doing its job.
Common mistakes
- Counting by person. As soon as someone knows their questions are counted under their name, they stop asking and guess instead. The number drops and the risk rises. On this point, read can your manager read your questions to an assistant?.
- Taking the drop as proof. Without an observed routine or recorded mistakes, a falling line may only mean the team has gone quiet.
- Measuring everything. Two routines, four signs. Any more and nobody keeps the sheet after week three.
- Leaving the sheet to the manager alone. The mentor is at the station and hears the questions. The manager reads the line.
- Drawing conclusions in week two. The early days measure novelty, not learning.
- Letting the tool decide who is independent. Sign-off stays human: a mentor who saw the task done, a date, a signature.
What to measure
- The number of repeated questions per procedure, week after week. It should fall.
- The time between starting on a station and the first opening or closing done alone and signed off, compared from one starter to the next.
- The share of "why" questions among those recorded. It should rise.
Where Tsuno fits in
In Tsuno Compétences, skill levels are validated and dated by a manager, and the "can train others" level is earned through on-the-job training. If closing is one of the station's skills, the date it was signed off already exists, attributed to someone who saw it done. Opening and closing checklists are done on a tablet, with their history. The pathway combines theory, hands-on practice, observation in real conditions and assessment.
Tsuno San, the intelligent voice assistant for frontline teams, gives people at their station access to the procedures, tasks, checklists and routine checks the team has written. It assesses nobody and does not decide who is independent. It acts within the person's own rights and carries out nothing without their confirmation. Its voice interaction, San Voice, is available as a demo.
What a good tool should also do: show, procedure by procedure, how often it was asked about, without ever saying by whom, and give the reason for a step when asked. An assistant people need less and less for the same task is the goal we are building San towards. We won't describe it as done before we have checked it in the kitchen. Until then, Fadila's sheet of graph paper works with any tool.
Further reading
To decide whether a tool deserves its place before you sign, see gadget or tool? The 60-day test, and on who does what, the mentor and the assistant: who does what in training. For a structured way of training on the job, read learning on the job: AFEST and a workstation assistant. And to know when a repeated instruction points to a system problem, how many times should you repeat an instruction before changing the system?. The framework is in the pillar page AI for frontline managers: analyse, advise, reassure. On what stays with people, read passing on a skill: what no AI will replace.
Key takeaways
An assistant builds independence if, on the same procedure, questions fall and then turn into "why" questions, until the routine is done alone and signed off by a mentor. Count by procedure and by station, never by person: a question asked to do the job well should never become a file.
Frequently asked questions
How can you tell whether a workplace assistant is teaching your team or making it dependent?
Track four signs over several weeks: repeated questions on the same procedure should fall, the time before an opening or closing is done alone and signed off by a mentor should get shorter, questions should shift from how to why, and mistakes observed at the station should go down.
Should you track the questions each employee asks?
No. A question asked in order to do the job well should never become a file on that person. Count questions by procedure and by station, with no names, to find out which sheet or which step is causing trouble.
How long does it take to see whether a team is learning with a new tool?
Eight weeks give a reasonable reading on daily routines such as opening and closing. The first two weeks are mostly a baseline: draw no conclusions before week four.
Why are why questions a good sign?
Because people only ask why once they know how. A question about the reason for a step shows someone trying to understand and adapt, not just to follow instructions.
What should you do if questions don't fall after several weeks?
Look at the procedure first: a vague step, a missing reason, a method that changes with each trainer. Only then look at the tool, if it gives the step without ever explaining it. The signal is about the procedure, not about people.
Does a drop in questions prove the team has learned?
Not on its own. Questions can drop because people no longer dare to ask. Cross-check it with mistakes observed at the station and with a routine done alone, then signed off by a mentor.