Why good organisations end up making people do work their systems should already be doing.

Perhaps unsurprisingly, I was working on ORAC, my engineering laboratory, trying to remove a source of irritation that had gradually crept into the engineering process.

Nothing was broken, ChatGPT was doing what it does well and Codex was happily working away in the background. GitHub was keeping track of the engineering changes and ORAC's Engineering Knowledge Engine was doing exactly what it had been designed to do. Each component was behaving perfectly sensibly.

The software wasn't the source of my irritation.

I kept finding myself moving information between them. A prompt here, a status update there, a file upload, a result copied back into another conversation. None of it was particularly difficult and none of it took very long, but after a while I realised I wasn't really solving engineering problems. I was transporting information between systems that already knew perfectly well what to do with it once they received it.

It made me smile because, had I been designing the software from scratch, I know exactly how I would have solved that problem.

I'd have written an API.

Instead, the API was me.

It wasn't really about AI

The more I thought about it, the more familiar it felt. It was a pattern I'd seen emerge time and again as businesses grow.

Nobody deliberately designs an organisation where people spend their day carrying information from one place to another. These things evolve. A new system is introduced because it solves an immediate problem. Someone creates a spreadsheet because it's quicker than waiting for a proper integration. A report gets emailed because that's the fastest way to get it into the hands of the next person. Every individual decision is perfectly sensible because it removes today's pain.

Over time, though, those sensible decisions begin to interact with one another. Before anyone really notices, the organisation has developed an ecosystem where information spends almost as much time being moved around as it does being used.

I've always thought of these as little operational eddies.

Little operational eddies

In a river, an eddy forms where the flow changes direction and water starts circulating instead of moving downstream. Organisations develop their own versions. Information arrives, somebody copies it into another system, waits for an approval, reformats it, emails it on, uploads it somewhere else and eventually it continues its journey. Nothing has gone wrong, but very little has moved forward either.

Sometimes they become little operational black holes instead. Five minutes here. Ten minutes there. An email that didn't really need sending. A spreadsheet that nobody trusts but everyone updates anyway. A weekly task that nobody questions because "that's how we've always done it". Individually they barely register. Collectively they consume hours, days and eventually significant amounts of money.

The recent explosion of AI tools hasn't created this problem. If anything, it has simply made it easier to spot.

As more organisations experiment with ChatGPT, Copilot, Claude, Gemini and countless specialist AI tools, the same pattern is appearing again. People copy prompts, context, files and results between otherwise capable systems simply because those systems don't yet exchange information in the way the workflow really needs.

I've started referring to that particular version of the problem as "Prompt Ferrying". Not because prompts are unique, but because they happen to be today's most visible example of a much older organisational pattern.

Ten years ago it might have been spreadsheets. Today it's prompts. Tomorrow it'll be something else.

The real question is whether we're asking our most capable people to spend their time where they add the greatest value.

People are remarkably good at applying judgement, spotting risk, weighing evidence and making difficult decisions. It's difficult to believe that carrying information between systems is where they add the greatest value.

If your most experienced people spend a significant part of their day moving information between systems, you've probably automated many of the inexpensive tasks while leaving the most expensive part of the process untouched.

That isn't a criticism of the technology, and it certainly isn't a criticism of the people. Every workaround almost certainly made perfect sense when it was introduced. Most organisations evolve this way because they're busy solving today's problems rather than redesigning yesterday's processes.

The challenge is recognising when all those perfectly sensible decisions have accumulated into something much larger.

The question worth asking

Perhaps the next time you watch a process working its way through your organisation, don't just ask whether it works.

Ask how many times a person has to carry information from one place to another simply because the surrounding systems don't.

If the answer is "quite a few", don't start by looking for another tool.

Start by asking why your organisation evolved that way in the first place. You may find the systems aren't really the problem at all. They're simply reflecting the way the organisation has evolved.

And if, somewhere along the line, one of your best people became the API... perhaps it's time to retire them from that job?