The real skill isn’t AI. It’s seeing what can be automated.

AI isn't replacing people. But the people who learn to see and automate the work around them are already pulling ahead, and the first step is smaller than most expect.

This piece draws on more than two years of automating processes inside a large company. It's practical experience, not research.

I did not start as an expert

When process automation first landed on my desk, I knew nothing about it. I hadn't taken a course, and no mentor handed me a method. I learned one problem at a time and made plenty of mistakes.

That shapes how I teach. In front of beginners, I'm not talking about something I always knew. I remember exactly what it's like to stare at a blank workflow with no idea where to begin, and that memory tells me which explanations work and which only sound clever.

More than two years on, I'm responsible for automating processes at work. Some of it is plain everyday routine. Some of it is new solutions that make a working day noticeably more efficient. One habit connects all of it, and it's the subject of this article.

Most work is a sequence of steps

I look at work as a sequence of steps and ask, for each one, whether a person really needs to do it.

A form gets filled in twice. A document waits for the next signature. An approval sits in an inbox because nobody owns the reminder. Once you start looking, you find these everywhere. They rarely register as problems, because everyone has accepted them as routine.

You don't need to be a programmer to spot them. You need to be curious about your own work and a little impatient with it. In my experience, those two traits matter more than any tool.

One app, around 1,700 colleagues

The project I'm best known for at DHL is an app for employment reference letters. It replaced an existing SAP application, and around 1,700 colleagues use it today. I built it with the Microsoft Power Platform.

Nobody had asked for a replacement. The SAP solution worked, in the sense that a result came out at the end. But it was slow, and a reference letter passed through many hands before it was done. People had learned to live with that.

The new app does the same job, only faster and simpler. It saves around 20 to 30 minutes of pure processing time per letter. A letter can also be created and used immediately, without waiting for it to travel from one desk to the next.

I want to be precise here: this is an ordinary business application built to solve a practical problem. What made the difference was one simple question. Why does this take so many steps?

What colleagues actually noticed

I expected people to talk about speed. Some did. The reaction I remember best was a different one: colleagues were glad the app is so simple.

That says something about the tools most of us put up with. We tolerate clumsy software for years, until we stop noticing it. When a tool that simply makes sense replaces it, the relief is bigger than the time saved.

It also changed the conversation. Once 1,700 people had seen that a tool they'd tolerated for years could be replaced by something better, the question shifted from "can this be improved?" to "what else can?"

AI is not replacing people, but it is sorting them

I don't believe AI will replace most people in the coming years. That's exactly why the shift is easy to miss. Nothing forces anyone to react.

There's no dramatic moment and no announcement. The change is slow. Colleagues who experiment build skills. They finish tasks faster, take on bigger problems, and become the people others ask. Those who wait eventually notice that the work has moved on without them.

I see this in my own environment. People who engage with AI and automation get better at their jobs. People who don't stay where they are while the baseline rises around them. Nobody gets pushed out. They just fall behind.

That's also why I don't see AI as a threat to those who build and improve processes. It multiplies what they do. A model can read an unstructured document or draft a text, but it only becomes useful inside a process someone has designed: where the input comes from, who checks the result, where the output goes next. Whoever designs that process decides how much value AI creates in a team.

Why I reach out to people

I approach colleagues directly. I'm highly engaged at work and follow AI developments closely, so I'd rather bring a topic to my colleagues than wait for them to come to it.

This isn't just personality. It's a method. Automation depends on knowing how work really happens, and you can't learn that from a desk alone. The details live in the heads of the people doing the work: the step they skip, the workaround they invented, the field they always fill in the same way. You only learn these things by asking.

I also share what I find. A good idea in one team can become the standard in ten, but only if someone carries it across. I try to be that person.

Teaching beginners, and why managers are next

I teach beginners today. That feels right, because the first step is where most people get stuck and where a good explanation changes the most.

I don't give them a list of features. I give them the habit described above: see the steps, question each one, build a small version, show it to the people who would use it. When that works once, people get curious about the rest.

I hope to bring this to managers soon. They decide which processes get attention and which tools are allowed, so a manager who sees the potential can unlock it for a whole team. It's a different audience with different questions, and I'm looking forward to learning how to speak to it.

It does not stop at the office door

I think the same way at home. Outside work I'm at the gym, on the golf course or out running, and I don't want to spend my little free time on tedious chores. So I automate them.

I mention it because it shows what the skill really is. It isn't tied to a platform or an employer. It's a way of looking at repeated tasks and refusing to accept that they have to be done by hand.

Where to start

Pick one task that costs you time or patience every week. Write down its steps and remove the ones that exist only out of habit. Then see whether a simple app or workflow can do the rest, and where a model could help with the reading or writing. Build a small version and show it to the people who would use it.

If it works, others will ask for the same. That's how an app for one team becomes an app for 1,700 people.

A note on sources

This piece rests on my own experience, not on external studies. The figures for the reference-letter app, the number of users and the time saved per letter, are my own account of the project. The view that AI sorts people rather than replaces them is my own assessment.