Here's how to think about it: the novice carpenter makes the thing. The skilled carpenter makes the thing that makes the thing — the jig, the fixture that lets the same cut happen reliably without a skilled steady hand guiding every pass.
The conceiving and making of the jig is a good example of an abstraction of work.
It's not even close to a new idea. Recipes do the same job for cooking. Templates and checklists do it for office work. A list of instructions to get something done. A macro does it for repetitive clicks.
In pretty much every skilled endeavour, at some point someone stops doing the task directly and starts building something that lets the task get done reliably by someone — or something — else.
Working with AI is basically forcing that move on everyone. Instead of doing the task yourself, you're stepping back and helping AI do the task — writing good enough instructions, building the right structure, so AI can reproduce the result reliably, at a scale no individual could match by hand. Get the instructions wrong, and AI reproduces the mistake at the same scale. That's the trade-off.
Business analysts, process improvement specialists, and countless others have been professional abstractors for decades in modern business settings. Abstracting work was their day job. Now it's everyone's job.
Not new, just on a new scale — impacting more people. This used to be something a few experienced people did occasionally. Now it's much of the job for many people.
And it's not something you do once and move on from. Working with AI, in an AI Business 2.0, means operating significantly at the abstracted level as the normal shape of the job — not a transition project with an end date, but the ongoing substance of the work itself. The jig doesn't get built once and left alone; it keeps getting rebuilt, because what AI can do keeps changing, and so does what's worth abstracting next. This isn't a phase people pass through on the way to somewhere else. It's what the job is now.
And once you've abstracted the work into something reusable, the people who used to do that work by hand start looking redundant to each other, not just to the machine. Ten accountants doing basically the same job never used to look like duplication — they looked like a department. Once the job gets written down, templated, directed, then ten copies of the same capability aren't as inherently useful as they were.
Full, reductionist dedupe — stripping a function down to what's sometimes worth calling its Minimum Viable Agentic Business Unit, the bare floor that still functions — is one path, and sometimes it's the right one. But it's the crudest first-order response, not the only one. There's a more strategic, multi-dimensional version of this decision, and it gets its own full treatment in the next piece in this set, The Deduplication of Labour.
This is the same thing sometimes called "the dedupe" — just made very concrete and human here: a room full of colleagues, one abstraction away from that room getting a lot smaller.
But deduplication is a crude view of what to do as a result of AI Labour. An unconsidered deduplication is a collapse.
That collapse has many actual costs, beyond the obvious headcount story. You lose built-in redundancy. Ten people weren't just ten units of output — they were also cover for illness, a second pair of eyes catching mistakes, institutional memory spread across more than one or two heads. Compress the team and all of that gets compressed too, unless someone deliberately decides to keep some of it. There's a real difference between duplication that was always waste and redundancy that was always a deliberate buffer — worth knowing which is which before you cut either.
You break the talent pipeline. The traditional way people became senior was by doing the job long enough to build the judgment to do it without instructions, and eventually enough judgment to direct someone else doing it. If that repetitive tier is exactly what gets abstracted away first, that path just doesn't really exist anymore. Somebody still needs to grow into tomorrow's expert judgment. Nobody's fully worked out yet what replaces the apprenticeship that used to build it.
And accountability doesn't automatically survive the collapse either. Ten people reporting to a manager had an implicit answer to "who's accountable for this" built into the org chart. Two or three people plus whatever AI does the rest don't inherit that answer automatically — someone has to decide it on purpose, or it defaults to whoever's left standing, which isn't the same thing as actually deciding.
None of this is really about carpentry, or accounting, or any one profession. It's just what happens, structurally, any time "making the thing" turns into "making the thing that makes the thing" — which, in an AI business, ends up being most jobs eventually.