The Skill Premium Collapse

The chainsaw was the first warning.

Before chainsaws, logging was a craft. A skilled logger felled three times more trees a day than an unskilled one, and earned three times more. The "skill premium" was real — it paid for years of practice and physical risk.

Then chainsaws arrived in the 1960s. Any unskilled worker with one could fell trees almost as fast as a master. The skill premium collapsed in a single generation. And then it stayed flat.

GPS did the same thing to taxi drivers in the 1990s. A London cabbie's "Knowledge" — three years of memorizing 25,000 streets — was the moat that justified the premium fare. GPS turned that moat into a $20 phone app.

Same shape on the chart. One sharp drop. Then a new equilibrium where any driver with a phone could compete.

For 100 years of industrial history, this was the pattern. Tools commoditize skills. Skill premium collapses. New equilibrium. Workers adapt to new tasks. The cycle restarts.

AI doesn't follow that shape.

Look at the third curve. It doesn't drop then plateau — it keeps falling. Every month the gap between a skilled and unskilled knowledge worker shrinks a little more. The curve has no obvious bottom.

Why? Because AI does something neither chainsaws nor GPS could do: it learns from every use.

A 2023 Harvard Business School study at a top consulting firm tested this directly.

Without AI, the gap between top-performing consultants and bottom-performing ones was about 20%. With AI augmentation? 4%.

That's a 5× compression in one tool generation. The "top talent" — the person you paid the skill premium for — was suddenly only marginally better than the median hire. And the median hire was suddenly only marginally cheaper to source.

This is the absorption flywheel.

Every time an experienced worker uses an AI tool, the tool watches what they do. The tool absorbs some fraction of their judgment. The next worker who uses the tool starts with that judgment built in. The gap between expert and novice shrinks. The tool gets better. Loop.

Chainsaws never got smarter. GPS never got smarter. AI gets smarter every day, and every worker who uses it makes it smarter still.

The economic consequence is subtle but important: the worker still does the task. They just can't charge for it anymore.

The value isn't being automated away. It's being relocated. It moves out of the worker's hands and into the tool's. Whoever owns the tool — OpenAI, Anthropic, the vendor whose AI absorbed your team's judgment — captures the skill premium that used to live in payroll.

The worker remains. The premium leaves.

Reskilling is a losing game in a system that's already changed. Every new skill you learn is a candidate to be absorbed by the next flywheel.

— From Reshuffle, Chapter 4–5

The real move isn't to learn more tasks AI can't do yet. It's to identify the new constraints the system exposes — and rebundle your role around managing them.

That's the sommelier's playbook, the anesthesiologist's playbook, the senior consultant's playbook. We'll meet them next.

Sangeet on this in Chapters 4–5 ↗