For as long as business has existed, there's been one dependable rule: to produce more, you hire more people. Ten accountants produce roughly ten accountants' worth of accounting. Add an eleventh, get more output. That relationship — more labour in, more output out — has been close to a law of business planning. It's baked into how companies budget, how economists model growth, how "scaling" gets talked about in pretty much every industry.
That rule is quietly breaking, and the abstraction of work is why.
Here's the mechanism, in short: once work gets abstracted — jigged, templated, handed to AI to execute — the people who used to do that work by hand stop looking like a team and start looking like duplicates of each other. Ten accountants doing basically the same job never used to read as duplication; they read as a department. The moment the job itself gets captured well enough for AI to run it, the ten start looking like ten copies of one capability. That's deduplication: not layoffs as a cost-cutting decision, but the plain fact that duplicated human skill, once captured once, doesn't need to exist ten times over anymore.
It's worth being precise about what deduplication isn't. It isn't the same thing as the abstraction of work — abstraction is about *how* the work gets done differently; deduplication is what happens to the *people* once it has. And it specifically isn't about cutting things that were never duplicated in the first place. A lot of what looks like a big team is actually deliberate redundancy: cover for when someone's sick, a second pair of eyes on a judgment call, contestability — someone else who can catch it if the first person, or the AI, gets it wrong. None of that is duplication. It's insurance, and it's usually worth keeping. The distinction that matters is between the duplication that was always waste and the redundancy that was always on purpose. Dedupe the first. Be careful with the second.
**What a business actually does with what gets freed up**
Deduplication frees up something — call it a dividend: human capital that isn't needed to do what it used to do. What a business does with that dividend isn't automatic, and it's really a strategic choice with at least three shapes:
- **Cut.** Bank the freed capacity as cost savings. Fewer people, same output, lower cost. This is the obvious move, and for a lot of commodity, low-margin work, it's the right one — if the market for what you do isn't elastic (people don't want more of it just because it got cheaper), there's nowhere else for the dividend to go.
- **Grow.** Keep the same people, get more output. If demand for what you do is elastic — more of it gets bought if you can supply more — the same team, freed from duplicated grunt work, can serve a bigger market instead of a smaller one.
- **Deepen.** Keep the same output, but make it more robust: more depth, more checking, more resilience, more of the deliberate redundancy mentioned above. Not more volume, better volume.
Most real strategies are some mix of the three. Businesses under real margin pressure lean Cut. Businesses with room to expand lean Grow. Businesses in high-stakes, low-margin-for-error categories often should lean Deepen, even when Cut looks tempting.
One more layer of precision worth adding, because Cut can sound cruder than it needs to be. Full reductionist dedupe — stripping a function down to whatever's worth calling its Minimum Viable Agentic Business Unit, the bare floor that still works — is one legitimate version of Cut, and sometimes it's the right one. But it's the crudest version, not the default. Even businesses leaning Cut usually do better retaining some capital deliberately: for risk mitigation (someone who can catch what AI gets wrong), for future expansion of the function (capacity that's cheaper to keep than to rebuild later), as a general buffer, or for resisting regression, where a deliberately retained ratio of human capital is specifically what keeps AI output from drifting toward generic over time. And where demand for the function is elastic, the dividend doesn't have to leave the building at all — it can be redirected rather than released, which is really Grow wearing Cut's clothes: headcount looks unchanged, but what it's doing underneath isn't.
**What this does to how businesses are put together**
Zoom out from one team and this changes the shape of organisations generally. A business used to be, in large part, a stack of departments — sizeable groups of people doing similar work, managed in layers designed to coordinate that similarity. Deduplication hollows out the middle of that stack. What's left, per function, increasingly looks like a small, complete unit — a person or two directing, a person or two augmented, AI doing the volume — rather than a department. Multiply that across every function that gets abstracted, and the org chart stops being a pyramid of similar specialists and starts being a network of small, complete units, each shaped roughly the same way. That's not a minor management fad. It's a different kind of business, built out of a different kind of building block.
**And zoomed out further, this is bigger than any one company**
If this is happening inside one accounting firm, it's genuinely a management story. If it's happening across every accounting firm at once — which is exactly what "specialties collapse" implies — it stops being a management story and becomes a labour-market story. An entire category of jobs can shrink in aggregate demand, not because any one company decided to cut, but because the underlying task got captured well enough that duplicating it across a whole profession stopped making sense. That's the same underlying flip that drains demand from a single business when the market's needs get met differently — except here it's draining from an entire category of *labour*, across many businesses, rather than draining from any single business's customers. Same flip; a different, wider surface for it to show up on.
**The old law and the new one**
Which brings us to the sharpest way to say all this. The old production logic was simple: more output needs more people. It's such a basic assumption that most business planning doesn't even state it out loud — it's just there, underneath the hiring plan, the headcount forecast, the org chart.
The new logic is close to the opposite: more output doesn't need more people. It needs more tokens, and better direction. Instead of adding headcount to scale, a business increasingly scales by marginally increasing its AI budget, and investing in the judgment that directs it well. Labour, which used to be the variable that flexed to meet demand, is increasingly the fixed part of the equation; AI capacity and direction quality are the new variable.
That's not a small tweak to how businesses plan. It's a different production function — the economists' term for the relationship between inputs and output — and most businesses are still planning against the old one.