Against unnecessary work

Debates about AI and work tend to assume the work itself is fixed: a lump of tasks, to be divided between people and machines. Walk through any organization and the assumption dissolves. A remarkable fraction of working time goes to work that exists only because systems fail to connect — re-keying data between tools, reconciling copies of the same truth, writing status reports that describe state a machine already holds, attending meetings that exist to compensate for documents nobody could find.

This category deserves a name and a policy. The name is unnecessary work: effort that adds nothing a well-built system could not have made unnecessary. The policy is that it should be eliminated, not optimized. Automating a reconciliation workflow is building a faster machine for a job that should not exist; connecting the two systems so they cannot diverge is the real saving, and it is permanent.

The distinction matters for AI in particular, because assistants are spectacularly good at performing unnecessary work — drafting the redundant report in seconds, smoothing the friction just enough that nobody asks why it is there. A world where machines do the unnecessary work faster is not a world with less of it; historically, lowering the cost of a wasteful process produces more of the process.

So we hold a simple discipline: before building a tool to do a piece of work, ask whether the work should survive at all. Necessary work deserves better tools. Unnecessary work deserves abolition. Confusing the two is how a civilization ends up busier after every labor-saving invention it adopts.