What we hand over is not our work.

It’s three in the morning in a control room. The screen that tells everyone what the network is doing freezes.

Nothing has failed yet. The pumps are still running, the water is still moving. But for the first time in years, nobody in the room is being told what to look at.

There is a pause that everyone in that room will remember. Someone reaches for the phone to call the supplier. Someone else looks at the oldest person on shift. And in that pause, the organisation finds out something it didn’t know it had agreed to: that it no longer fully understands its own work.

I’ve seen versions of that pause in almost every organisation I’ve worked with. It isn’t always a frozen screen. Sometimes it’s a contractor leaving, a system being switched off, or a simple question nobody in the building can answer without opening a tool. It always arrives the same way. Quietly, and too late to have been a decision.

Borrowed, not bought

Every tool we bring in makes an offer. Save an hour a day. Spot the pattern you’d miss. Let the supplier worry about the hard part. The offers are genuine, and saying yes is usually the right call.

But every offer has a second half that isn’t in the contract. The tool takes the work, and the practice of doing the work goes with it. Skills that aren’t used fade. Knowledge that lives in a system stops living in people. Judgement about when the tool is wrong depends on knowing how to do without it, and that is exactly the thing we’ve stopped doing.

None of this feels like a loss while the support is there. That’s what makes it borrowed rather than owned. You don’t notice a capability is on loan until someone asks for it back.

What we hand over is not our work. It is our capacity to do without help.

The irony nobody planned

This isn’t new, and it isn’t really about AI. In 1983 the engineering psychologist Lisanne Bainbridge wrote a short paper called Ironies of Automation. Her point was simple and uncomfortable. Automation takes over the routine parts of a job and leaves people with the rare, difficult moments: the failures, the edge cases, the situations nobody designed for. Those are precisely the moments that need the deepest skill, and the people now facing them get the least practice. The more we automate, she argued, the more crucial the human contribution becomes, and the harder it is to keep.

Aviation learned this the hard way. In 1997 Warren Vanderburgh, a training captain at American Airlines, gave a talk that became known as Children of the Magenta Line, after the line on the flight display that pilots had learned to follow. His warning was that crews were becoming so fluent in managing the automation that they were losing the instinct to simply fly the aircraft when the automation was the problem. Years later, in 2013, the US aviation regulator formally urged airlines to give pilots more chances to fly by hand.

Most organisations haven’t had their equivalent moment yet. They are adopting tools faster than ever, and AI is the fastest of all. Drafts that used to be written by junior staff are now generated. The judgement those juniors would have built by writing badly, being corrected, and writing again has nowhere to come from. Nobody chose to stop growing judgement. It just stopped being part of anyone’s job.

How to tell it’s happening

Borrowed capability rarely announces itself. These are the signs I look for:

• When a key system is down, work stops rather than slows.
• A supplier or contractor knows your operation better than your own people do.
• You sign off work you could not realistically challenge.
• The people who could do it by hand are near retirement, and nobody is learning from them.
• New starters learn the tool, not the job.
• Nobody can say which capabilities you have decided to keep, because nobody has decided.

That last one matters most. Borrowing is fine. Borrowing without knowing what you’ve borrowed is the problem.

What to do about it

The answer isn’t to refuse the tools. That would be the old trap of opposing rather than offering something better. The answer is to borrow on purpose.

Run a fire drill for capability. Choose one routine task, switch the tool off for a few hours with a safe fallback in place, and do the work by hand. Then talk honestly about what was missing. It turns an abstract risk into something a team has felt, without anyone arguing about whether the tool is good.

Write a keep list. Decide, explicitly, which capabilities must stay in your people even if a tool can do them. Not everything. The few things you would need on the worst day.

Keep the hard ten per cent by hand. If skills fade without practice, build practice in. Pair experienced people with newer ones. Rotate the manual work. Treat it like a fire alarm test, not an inefficiency.

Ask your suppliers to teach, not just deliver. If the logic of a system lives only in the supplier’s head, ask for it to be written down and understood in-house. That isn’t distrust. It’s the difference between renting a capability and owning one.

Grow judgement deliberately. If AI now does the first draft, someone still needs to learn what a good draft is. Build that learning in, or accept that in five years nobody will be able to tell.

Where this connects

Borrowed capability is one of twelve ideas on the Oppositism map. It sits in the first phase, See, because it is invisible until you look for it. It leads naturally to Decisions nobody made, because nobody decided to lose the skill, and to Unlearning, because choosing what to keep also means choosing what to let go. And it comes back, like everything, to Ground truth: the work as it is actually done, by the people who actually do it.

If you only take one question away, let it be this: what have we borrowed that we’ve started to think we own?