The Three Kinds of Work
Work no longer divides in two. There is work that got cheap and can be handed off. There is work that is still yours: knowing where to start, knowing when the question is wrong, knowing which real case matters more than a tidy theory. And there is a third kind, thin but real: catching what the machine got wrong. The third kind is not the old work, and it is not free.
Framework 2 of 10 · Chapter 2: What's Getting Cheaper
The familiar split is between work a machine can do and work it cannot. It is not wrong, and it is not sufficient, because it leaves out the work the machine creates.
Work that got cheap. Boilerplate. Test setup. A validation layer. Schema plumbing. Real work, formerly yours, now available at a much lower price.
Work that did not get cheap. Knowing where to start. Knowing when the question itself is wrong. Knowing that one real case in production outweighs a clean theory about all of them.
Catching. The machine produces something plausible and someone has to notice what is wrong with it. This is new work. It is thin per instance and it accumulates, and almost nobody budgets for it.
When it applies
At the end of any week where the volume of output went up and you cannot say whether the quality did. Also before any estimate, because the three kinds do not scale the same way and pricing them as one number is how estimates go wrong now.
How to run it
- Sort last week's work into the three kinds. Actual items, not categories of item.
- Then run the harder pass. Look at everything you approved, and ask which of it you reviewed.
Approved and reviewed are not the same word. Four lines you understood and twenty-six that looked fine is a normal week now, and the distance between those two words is where the risk of this decade lives.
Where it fails
You price catching at zero. It is the most commonly unbudgeted work in engineering right now, and the cost surfaces as an incident rather than as a line item.
You treat the boundaries as fixed. The line between the first two kinds moves. Sorting once and trusting the sort is its own failure, which is what the handoff line is for.
Practice
Sort last week's work into the three kinds. Then list everything you approved and mark what you actually reviewed. Do not fix anything yet. The list is the finding.