Why restaurant employees quit the job you designed

I've watched people at ChurroMania become incredible operators. They can tell when a guest needs attention, when someone on the team is ready for more responsibility, and when the store has stopped feeling the way it should. Some have developed those abilities over years, even decades with the brand.

Then we ruin it: we give them reports to reconcile, schedules to assemble, orders to check, and a collection of systems that all need something from them. Being good at the job earns them more of the work that keeps them away from the part they're good at.

Black Box Intelligence put limited-service hourly turnover at 110% over the twelve months ending in the third quarter of 2025.

There are reasons for that beyond the work itself. Pay matters. So do predictable hours and the person you report to... But we think we’ve all treated the design of the restaurant job as something fixed while trying to improve everything around it.

The people who used to be computers

In the 1790s, the French mathematician Gaspard de Prony organized the production of an enormous set of mathematical tables. There was too much calculation for a small group of mathematicians to complete themselves, so he divided the work.

At the top, a small group selected the mathematical formulas. A second group converted those formulas into numerical instructions and checked the results. At the bottom, dozens of people carried out addition and subtraction. The human calculators included former hairdressers who had lost work after the French Revolution. They didn't need to understand the whole mathematical problem to perform their part accurately.

Babbage described the arrangement in On the Economy of Machinery and Manufactures. His account identifies five or six mathematicians in the first group, seven or eight people in the second, and sixty to eighty calculators in the third. He then describes what a technological calculating engine could become: "a substitute for the whole of the third section of computers."

The machine had a job waiting for it: People had already separated the thinking, the translation, and the repetitive execution.

That was an effective way to produce mathematical tables, but it also gives us an interesting way to look at a restaurant manager's day. How much of it requires someone to understand the whole operation, and how much consists of small steps that someone else could specify?

These are categories of work, not categories of people. The person doing arithmetic could be capable of much more than arithmetic. The person exporting a report may be the best untapped leader in the building.

What we ask our best people to do

Any brand can write on their website that hospitality matters to them, but it’s the manager that has to make it happen with the team working with them that afternoon.

That means knowing how to coach someone without embarrassing them in front of everyone. It means noticing when a new team member has memorized the process but doesn't understand the guest. It means making a busy store feel welcoming without pretending it isn't overwhelming.

I think of that as the ‘translation’ layer. The manager takes the intentions of the brand and turns them into something other people can understand, perform, and feel.

We need more of that work. Yet the same person may spend the morning collecting information from systems that already have it, arranging it into another format, and checking whether a routine task was completed. Although those things matter, they don't all require the person we currently assign to them.

Reporting is certainly necessary. So, of course, is a reliable labor plan. Supplier orders need to be right. None of this becomes optional just because we'd rather be talking to guests, so the opportunity becomes giving more of the repeatable execution to technology that can actually complete it.

I was at the QSR Evolution conference recently, and there was an idea circling around very few operators that interested me a lot. I’m going to paraphrase it down to: “how can a brand mask ‘functionality’ with hospitality”. In other words, the restaurant has to serve its function optimally (speed, quality, consistency, etc.) but a guest shouldn't feel like another task passing through a system. The Wall Street Journal recently reported that major fast-food chains are reinforcing hospitality and service training now more than ever. ‘A return back to humans’.

When automation takes reporting and scheduling off a manager's plate, we want that time spent coaching the team and looking after guests. If we simply fill those hours with more administration, the manager still has the same kind of job. A job that’s easy to want to get away from.

Giving an assignment instead of another tool

My father has used A Message to Garcia as a lesson in taking responsibility ever since I can remember. Elbert Hubbard's essay describes the ideal team member who receives an assignment, works out what is missing, and delivers the result, no questions asked; the textbook definition of agency. It was published in 1899, 127 years ago.

I now keep a version of that doctrine in all my AI agent's instructions. Find the information and check the result, ‘trust but verify’. If an approach fails, try a different one until it succeeds. Whatever the blockers, the ultimate goal is to bring back the finished work rather than another list of things for me to do.

Of course, an instruction file doesn't make a system infallible. It still needs access, boundaries, and a way to recognize when a decision belongs to a person, but it changes what I can reasonably ask the technology to take responsibility for. Modern agent systems can choose steps and use tools dynamically, rather than only follow a fixed sequence, although errors can still accumulate and verification remains necessary.

That’s useful in a restaurant because a tool that saves one task but creates several new ones hasn't necessarily improved the manager's day. The ideal is to delegate a piece of work without becoming its full-time supervisor.

The job after the work moves

This is where an operator has to make a choice. Removing administration doesn't automatically give an employee a better job to perform. We could very well fill the space with more administration, increase their workload, or cut their hours. We could, however, also give them the authority to improve the experience and give them the time to get better at doing it.

The latter is the opportunity that we became interested in enabling.

A manager should be aware of what responsibilities replace the repetitive work: coaching a shift leader, developing a stronger team, and improving how guests experience the product, among many others. They should have the authority and support to do it, rather than being told to show initiative while every decision still requires permission.

What happens in this new environment is that more capable people want to stay in this new version of the job. We should measure whether they do, and we should be willing to admit when pay, staffing, or leadership is still driving them away.

Babbage's pyramid helped explain how to divide work. We’ve spent a long, long time putting people into the repetitive execution layer. We want to give more of that work to technology. We'd like to see what our best operators can become when we let them use the rest of their abilities.