The skill premium is what rare expertise used to earn. AI is compressing the gap between high- and low-skilled output across knowledge work. The premium isn't gone everywhere yet - but it's going.
For a century, the policy answer to technology displacement was "reskill into the next thing." It worked when the next thing was a stable, scarce skill that paid a premium for a decade or two.
AI compresses the premium faster than reskilling cycles can keep up. The chainsaw cut the loggers' premium once. GPS did it to cab drivers once. AI does it continuously.
Skill premium exists because some output is scarce - only some workers can produce it. Earlier technologies tended to shift which skills were scarce. The premium moved but kept existing. Reskilling worked because the new scarce skill stayed scarce long enough to be worth learning.
AI is different. It doesn't just shift the scarce skill - it compresses the gap between high- and low-skilled output. The legal junior writes a brief that the AI-assisted paralegal can match. The senior analyst writes a report the AI-assisted business user can match. Every use trains the tool. Every training narrows the gap further.
The honest implication: reskilling alone doesn't restore the premium, because the new skill is also being commoditised while you learn it. The strategic move isn't a new skill. It's identifying the new constraint in your system - the friction AI hasn't (and may not) dissolve - and rebundling your work around resolving that.
Wine info is free. Sommeliers earn more than ever. Why?
By every task-centric measure, the sommelier should be extinct. Any diner can pull up tasting notes, vineyard histories, and pairing recommendations on their phone. The information that used to be the sommelier's edge is free.
Sommelier pay has risen, not fallen. The role rebundled around a different constraint: helping people feel confident in a moment of uncertainty. Curation. Story. Performance. The old skill (memorising vineyard data) lost its premium. The new constraint (decision confidence under social stakes) earned a higher one.
The sommelier didn't reskill into more wine knowledge. They rebundled around the friction the new system had created. That's the playbook the rest of the knowledge economy hasn't internalised yet.
Reskilling is what you do after identifying the new constraint. Done in isolation, it's a treadmill that AI keeps accelerating.
Two years ago, the ability to write clean, idiomatic code in a popular framework was worth a meaningful premium. With AI coding assistants, the median output of a mid-skill developer is now comparable to what a senior engineer would have produced. The premium for "writes good code" is collapsing.
The new premium sits at "knows what to build" and "owns the consequences when it breaks in production." That's the constraint AI hasn't dissolved.
Long-form blog posts, ad copy, social content - all indistinguishable in quality from AI output, often better because AI doesn't get tired. The skill premium for "writes well" has collapsed inside two years.
New premium sits at "knows what to write about" - strategy, voice, brand judgement, the ability to decide what's worth saying. The reskilling answer ("learn to use ChatGPT") misses the real shift entirely.
What constraint in my industry's system hasn't AI dissolved - and can I credibly own it?
Reskilling without a constraint hypothesis is just running. Reskilling toward a constraint is strategy.
Stop chasing the skill. Start tracking the constraint.
An advisory engagement that redesigns the operating model and workforce strategy around the constraints AI doesn't dissolve - identifying which roles to bundle around which constraints, and how to allocate investment across people and AI capability.