When a business starts using AI, it almost always starts with people. Someone in finance begins asking ChatGPT to tidy up their reconciliation notes. Someone in marketing uses Claude to draft product descriptions. A manager starts summarising meetings with it. Nobody planned it. It just spread, one person and one task at a time.

That is a good thing. But it is only the first layer, and a business that stops there is building on sand.

I think about AI in an organisation as two layers: the personal layer and the systems layer. Knowing which one a piece of work belongs in is one of the most useful questions you can ask.

The personal layer

The personal layer is how people use AI within their own role, to improve their own work. Their chats, their prompts, their little tricks, the custom instructions they have tuned over months.

This layer is brilliant for discovery. The person doing the job knows where the pain is better than anyone, and when they have a capable AI to hand, they find improvements nobody in a meeting room would have thought of. The best AI ideas in most businesses are sitting in someone’s chat history right now.

But there is a catch. AI in the personal layer is still an expression of a person. Their way of asking, their context, their judgement about what good looks like, all of it lives in their account and in their head.

So when that person leaves, their improvements walk out the door with them. The clever routine that turned a two-day job into a two-hour one was never written down anywhere the business could find it. The next person starts from scratch, or worse, never learns it was possible.

It is the same problem businesses have always had with key people, except faster. AI makes a single person far more capable, which means there is far more to lose when they go.

The systems layer

The systems layer is where the good ideas need to end up. It is AI work that belongs to the business rather than to a person.

In practice that means three things.

It is documented. What the process does, why it does it that way, and what decisions were made along the way are written down in plain language, somewhere the business owns.

It is auditable. You can see what the AI did and on what information. If a number looks wrong, someone can trace where it came from, rather than asking whoever set it up to remember what they typed.

It is durable. It keeps working when someone goes on leave, changes role or leaves altogether. A new staff member, a consultant or an AI agent can pick it up and carry on.

The systems layer is less exciting than the personal layer. It is where things get written up, tested, handed over and kept running. But it is the difference between a business that has some people who are good with AI and a business that has actually got better.

Moving work from one layer to the other

The two layers are not rivals. The personal layer finds the ideas and the systems layer keeps them. The skill is in noticing when something has earned its promotion.

A few signs it is time:

  • Other people would benefit. If one person’s routine would save the whole team time, it should not live in one person’s chats.
  • The business depends on it. If a report, a customer reply or a decision now runs through someone’s AI routine, it has quietly become a business process and should be treated like one.
  • You could not explain it if they left. If the honest answer to “how does that get done?” is “ask Jo, she has a thing”, that is a personal-layer process doing a systems-layer job.

Promoting it does not have to be a big project. Often it is writing down the steps and the reasoning, moving the instructions from someone’s personal account into a shared place, and making sure the files it works on live in the business’s own systems rather than on one laptop.

Where to start

Encourage the personal layer. Give people good tools, let them experiment, and make it easy for them to share what they find. That is where your ideas will come from.

Then, every so often, ask one question: which of these would we lose if this person left tomorrow? Those are the ones to lift into the systems layer first.

If you want a hand working out which of your team’s AI habits deserve that treatment, and how to do it without a big programme, send me what you have.