Why we start on the floor, not in a slide deck

Every AI session I have sat through opened the same way. A slide about what a large language model is. A slide with a graph going up and to the right. A slide about how this changes everything.

Forty minutes in, a general manager who has been up since five is checking whether the delivery landed.

Nobody in a restaurant needs the theory. They need to stop doing one specific thing by hand, and everything that is not that is costing you the room.

The wrong first question

"What could we use AI for?"

Ask a room of managers that and you get silence, then somebody says chatbots, then somebody else says they read that it makes things up. Both are fair and neither goes anywhere.

The question does not fail because the room is not technical. It fails because it is abstract, and nobody thinks abstractly about their own job. They think in tasks.

The right one

"What did you do last week that you would be annoyed to do again?"

That one never gets silence. It gets a list, immediately, and the list is always specific: the Monday packet, the invoices, the schedule, the same email to the same vendor, counting the walk-in twice because the first count was wrong.

Every item on it is a candidate, and the person who named it already wants it gone. That is most of the battle, and no slide produces it.

Why the theory does not help

Knowing what a language model is does not tell you whether it can read your invoices. Knowing that it sometimes makes things up does not tell you which of your tasks that matters for.

What actually transfers is having done one. A manager who has taken their own invoice, got a wrong answer, worked out why, and got a right one has a working model of what the thing is good at that is worth more than any explanation.

They also have something else: a story about it being wrong. That is the part that stops a group over-trusting a tool six months later, and you cannot teach it from a slide.

What goes wrong when you start with the tool

You end up teaching a product instead of a skill.

The vendor changes the interface, or you switch systems, or the free tier changes, and everything the room learned evaporates because it was all about where the buttons were.

Start from the task and the skill survives the software. The manager who understands that the useful input is last year's same week, by daypart, by store, can carry that into whatever tool exists in two years. The one who memorised a menu path cannot.

The uncomfortable part

Starting on the floor means the first hour usually finds out that the answer is not AI.

The Monday packet does not need a language model. It needs three exports on a schedule and somebody to stop rebuilding it by hand. The invoices might need one; the schedule almost certainly does not.

That is a worse sales pitch and a better session. A room that watches you talk them out of two of their four ideas will believe you about the other two, and they will use those two.

What a good session actually looks like

Nobody watching one screen. Everybody in their own tools, on their own numbers, building the thing they named in the first ten minutes.

At the end, each person has something that works and a one-page printout of how it works. Not notes about a thing they might build later — the thing, working, on their data.

The test I hold myself to: if a line cook has to think about it during a rush, it was built wrong. That applies to the training too. If a manager has to remember a framework to use what they learned, it will not survive a busy Friday, and a busy Friday is the only environment that counts.

Who this is for

Anyone who has sat through the deck version and left with nothing.

You do not need a technical team. You need one job somebody is annoyed to do again, and four hours where nobody is paged.

---

Not sure which job to bring? Take the free check — eight questions, no email, and it names the two worth starting with in your building.