Don’t automate the workaround
Workarounds are evidence of underlying weaknesses or legitimate needs that must be understood before their compensating work is automated.
Organisations are full of workarounds. A spreadsheet exists because the main system does not quite provide what somebody needs. A telephone call happens because nobody trusts the delivery date. An additional check was introduced after something went wrong three years ago. A particular employee handles unusual cases because they have learned what works. Over time, these activities stop feeling like workarounds. They become the job. This matters when we automate. Imagine customer-service staff telephone Production every afternoon to check delivery dates. It would be easy to automate the transfer of that information.
But first ask why the telephone call exists. Perhaps Production is not updating the system. Perhaps different departments interpret the same date differently. Perhaps planning itself is unreliable. Or perhaps there is genuine uncertainty and the conversation between experienced colleagues contains useful judgement that a single date cannot represent. Until we understand which explanation is true, we do not know what to automate. This leads to a principle I think is worth remembering: Do not remove the compensation until you understand what it is compensating for. The same applies to AI.
A generative AI system might read several sources and reconcile conflicting information automatically. Useful. But why are the sources conflicting? An agent might chase missing information. Useful again. But why is the information repeatedly missing? Technology can become extremely effective at performing the compensating work organisations have accumulated over many years. That can deliver immediate productivity gains. It can also preserve the underlying conditions indefinitely. So when you discover a workaround, do not immediately eliminate it. Treat it as evidence. Ask what weakness, variation or legitimate need caused it to appear. Then decide what the work actually needs. Sometimes automation will be the right answer. Sometimes the workaround is pointing directly towards a better one.