The manager as messenger
Most management work is moving information from one part of a business to another. That is the part the tools now do. What is left is the judgement, and the judgement was the part worth paying for.
In September 2024 Andy Jassy told every senior team at Amazon to raise the ratio of individual contributors to managers by at least 15% within six months. Morgan Stanley counted 105,770 managers on the payroll and estimated that nearly 14,000 could go. The target was met by the end of March 2025, largely without redundancies. Teams were merged, and a number of managers went back to doing the work they used to oversee.
The tempting conclusion is that middle management is finished. Amazon kept most of its managers. What it cut was a portion of their week.
Most management work is moving information. A business grows past the point where one person can properly oversee five or six others, so it adds a layer, and later another. Each layer gathers what is happening below, decides what matters, passes it upward, and explains decisions from above on the way back down. The weekly report and the status meeting are coordination, performed by people, at salary. It is honest work, and it has been necessary since businesses first grew beyond one owner's direct supervision.
What has changed is that the records a business already produces can now be read by machine. Where decisions, plans and progress exist as documents, tickets and transcribed meetings, tools such as Microsoft's Copilot and Anthropic's Claude can compile what is happening across a team and answer questions about it on demand. Information that management layers were built to relay no longer needs a person to relay it.
Managers do more than carry messages, which is where cost-led delayering fails. They develop people, hold standards, notice who is struggling before it appears in any report, and have the conversations no software should attempt. Businesses that stripped out layers in the delayering fashion of the 1990s rediscovered this at some expense. The relaying work did not disappear, it passed to the people who stayed, and the experience that left with those managers proved slow and costly to replace.
For a business of £15m to £120m turnover the numbers are smaller than Amazon's, and looking at them is more useful. You employ two or three layers, not twelve, and removing one is rarely the answer. Look instead at how your managers spend their week. A manager on £60,000 who spends a portion of each week gathering and passing on status is buying message-carrying at a manager's salary, and message-carrying is the work the tools are learning to do. The rest of the week, the judgement and the people, is what you were paying for all along. The gain worth chasing is that portion of the week back: managers developing their people, and problems caught while they are still cheap to fix. Some businesses will take this as a reason to employ fewer managers. The better return is managers with the time to manage.
Jassy's ratio counted messengers. No ratio counts judgement, and judgement is the part worth keeping. Software will soon handle the relaying. It will not make the decisions.