The Original Version of the Work Appears Only Once
The last time I wrote about why AI training is failing for commercial workers, I ended up with a line that I knew would come back to haunt me. I said that a lot of this failure comes from building a course without anyone sitting down with a real student in the actual workflow, and that in the afternoon doing that changes what you’re building. A few people read that and asked the obvious follow-up. That’s right. How do you really run in the afternoon?
Good question. It sounds easy until you’re standing in a shop next to a finish specialist who you clearly wish wasn’t there. Here’s what I actually do, and the first thing I do is leave a clipboard in the car.
Go Where the Work Happens, Not the Conference Room
The first rule is that you don’t do this at the table. If you pull a salesperson into a meeting room and ask them to describe their day, you’ll get a clean, tidy version of their work that’s relegated to the audience. It will be accurate, and it will not be useless. People describe their work as a formal process, not a real one, in the same way that most of us would describe our morning routine as more calm and logical than it actually is.
So I go to the counter, or the bench, or the mixing station, and watch the work being done in front of me. The gap between what someone tells you they are doing and what they actually do is the whole game. You can only see it by being there when they do it.
Watch More Than You Ask
When I sit with someone, I look for times when they pause and decide something. The attorney takes the contractor’s question, thinks for a second, and pulls the product. The technician looks at the panel, looks at the specification sheet, and adjusts the measurement. Those breaks are where the tool helps or hinders, and is the only part of the job that should be designed around. The other thing is muscle memory that no AI tool will improve.
This is like learning a delivery route by riding shotgun with a driver who has run it for years. Printed directions take you to the right streets. Sitting next to them, you catch the little thing they do that the directions didn’t talk about: the corner they take the most because the truck always stops there, learning what they do in the loading bay before they commit to it. That judgment is a part you won’t find in the written version, and it’s an important part.
So I ask less than what people expect. When I ask, I ask about something I just watched. Why did you pull that one and not the other one? What were you looking at when you stood there? What happens when you get that wrong call? Those questions get real answers because they are tied to something concrete that just happened, not to an abstract version of the task.
One thing you should be able to tap into: people do less when they know they are being watched. The first time of any visit, you see a slightly cleaned up version of the work, the same way you would carefully drive with a passenger in the car. That’s why you stay. Give it long enough, and they forget you’re there, go back to the original rhythm, and start taking shortcuts and doing quick learning that they wouldn’t have shown in the first twenty minutes. A reliable version of the work is what you should be waiting for.
Questions That Get You Nothing
There are questions I have learned to stop asking. “Take me through your typical day,” you get the note. “Where do you think AI can help?” it gets you a blank look or a guess written by whatever they read last week, nothing tells you anything about their work. And “What’s frustrating about your job?” it makes people defensive, because it sounds like a set-up for a performance review.
The employee should feel like a guide showing you around, not a specimen under a glass. The fastest way to ruin that is to take notes when they say something that confirms what you already suspect. People are noticing. So I keep the notebook very closed and write things down in the back, in the car, while it’s still fresh.
It’s Afternoon, And That’s The Point
People ask how long this takes, hopefully I’ll say it’s a quick thing they can skip. It’s usually one good afternoon for each role you design. You won’t get to every role on the org chart, and pretending otherwise just sets you up to do them all wrong. So he makes the call: he spends an afternoon on two or three passages where the decision of the instrument has a lot of weight, and he accepts that he will know a little better about the rest. I’d rather be honest about that than suggest lessons from one bench are transferred cleanly. They don’t, which is the whole reason you went and sat down in the first place. You don’t use a research project or focus group or refer to a survey return as a spreadsheet of averages. You spend one afternoon watching one person doing real work, usually with your mouth shut.
What you’re going with are two or three choices where the tool can really help, and a few areas where it can slow down a skilled person. Build a module around that, and design the day the employee has. Skip it, and you’re guessing, and your finish line will be the only thing that looks good.
