AI can make bad work harder to see
AI can remove visible friction while making unnecessary or poorly designed work less likely to be questioned.
One of the obvious benefits of AI is that it can remove friction from work. A report that takes somebody a day to prepare might be produced in minutes. Information that has to be copied between systems might be transferred automatically. Emails can be drafted, documents summarised and administrative tasks completed with much less effort. All of that can be valuable. But there is another possibility worth considering: AI can make poorly designed work easier to tolerate. Suppose somebody spends half a day every month producing a management report. They gather information from several systems, correct inconsistencies, chase missing figures and assemble everything into a spreadsheet.
The obvious AI opportunity is to make that activity faster. But why does the report exist? Who uses it? What decisions does it support? Why does the information have to be assembled manually? Why are figures repeatedly checked? Which parts are actually read? If AI reduces the production time from four hours to ten minutes, we have certainly improved productivity. But if half the report serves no useful purpose, we have also made unnecessary work so cheap that nobody may question it again. The same applies elsewhere. An awkward information transfer can be automated. An unnecessary reconciliation can be performed by an agent. A badly designed approval process can be accelerated.
The friction disappears, while the reason for the friction remains. This matters because effort is sometimes evidence. When people repeatedly chase information, correct errors or spend hours producing something, the organisation eventually notices. AI can remove the symptom before anybody investigates the cause. That is why I think organisations need to use AI in two different ways. AI in the business performs work. AI on the business helps us understand the work. We should absolutely use AI to remove unnecessary effort. But before we make an activity dramatically easier to perform, it is worth asking a more fundamental question:
Should we be doing it at all?