Don't Automate the Workaround

One of the tempting things about AI is how quickly it can remove friction from existing work.

Someone copies information between two systems? Automate it.

A team spends hours turning data into a report? Generate it.

People repeatedly interpret the same document? Give an agent the task.

Often that will be exactly the right thing to do.

But there is another possibility.

The friction may be telling us something about the operating model.

A spreadsheet might exist because the formal system doesn’t match the real work.

A manual hand-off might exist because accountability is unclear.

A report might consume hours every month even though nobody can explain which decision it informs.

A workaround can be inefficient and still be useful evidence.

If we automate it too quickly, we may remove the inconvenience while preserving the reason the inconvenience existed.

Technology can make a good operating model dramatically better.

It can also make a bad operating model dramatically faster.

That feels particularly relevant now because the cost of automation is falling.

When changing a system required a large investment, there was usually pressure to understand the process before spending the money.

If we can now create an agent in days or hours, that friction disappears too.

Mostly, that’s a good thing.

But it means the discipline has to come from somewhere else.

Before asking what AI can automate, it may be worth asking why the work happens this way.

Should the task exist?

Could the information be captured once rather than moved repeatedly?

Is the process compensating for poor data somewhere else?

Does the approval add useful control, or is it simply inherited?

Is the workaround showing us something the service itself needs to change?

This isn’t an argument for a six-month process review before anybody builds a prototype.

Cheap experiments are valuable precisely because they let us learn quickly.

But the experiment should help us understand the work, not simply reproduce it more efficiently.

Sometimes the best use of AI will be to automate the workaround.

Sometimes it will be to help us discover that we no longer need the workaround at all.