Patch the right hole, in the right order
The hierarchy in plain language
The cleanest productivity instinct comes from a discipline well outside computing — an elimination-first ordering popularised in early-2000s lifestyle writing and grounded much earlier in operations research. The order is simple: first ask whether a task can be removed entirely, then whether it can be done by a machine, then whether it can be handed to another person. Most people invert this order. They take a process that exists out of inertia and ask which tool can do it faster. A faster way to do something that should not be done at all is one of the cleanest ways to lose a week.
Why elimination is the hardest step
Removing work is harder than automating it for two reasons. The task usually has a reporting line behind it — someone above asked for it, someone alongside depends on it, and the conversation to stop doing it is socially costly. The worker also has an identity wrapped around the task; being the person who runs the Monday report is part of how a role is defended. Tools cannot help here. Only a deliberate audit of what each artefact actually does for someone downstream — and a willingness to ask that person whether they still read it — can.
Automation as the second move
Once a task has survived elimination, the question becomes whether a deterministic script or a probabilistic model should run it. Deterministic scripts are cheaper, faster, and predictable; probabilistic models are slower, more expensive, and capable of handling input variety. A common mistake is reaching for a model on a task that a five-line script would handle, because the model is more impressive to demonstrate. The right rule is dull: use the simpler tool unless the task genuinely requires the more complicated one. The bias should sit against the bigger machine, not in favour of it.
Delegation comes last because it costs the most
Handing work to another person introduces a coordination cost that machines do not impose. Briefing, review, feedback cycles, and the risk of misalignment are overheads that scale with the number of humans in the loop. This is why the hierarchy ends with delegation rather than starting with it. People are precious. The right things for them to be doing are the things only a person can do well — judgement under genuine novelty, relationships, taste — not the things a tool could have handled for a small fraction of the salary.
Common failure mode — industrialising waste
The hierarchy exists because skipping any step produces a recognisable failure. Automate before eliminating, and you end up with a faster way of doing something pointless. Delegate before automating, and you have paid a salary to do something a machine would have done for pennies. Eliminate everything indiscriminately, and you find six weeks later that you have removed a task someone downstream silently relied on. The discipline is not the order in the abstract; it is the patience to walk through it one step at a time, with the people the work touches in the room.
Take a typical week and count how many recurring tasks you perform. Of those, which ones do you continue out of habit rather than because they currently matter to someone downstream?