Member-only story
A randomized study across 4,867 developers found that code-assistant access increased completed tasks by 26.08%. A Stanford payroll study found a roughly 16% relative employment decline for 22 to 25 year olds in highly AI-exposed occupations. Those are not two sides of one causal equation. They come from different datasets, populations, and questions.
Put them together anyway, because they identify the question that actually matters: when work gets cheaper to produce, does the market buy more of it, hire fewer people, lower prices, or stop training the people who would have learned to supervise the next generation of tools?
The evidence says AI is already replacing tasks and some paid deliverables. It does not yet show a detectable economy-wide unemployment shock. The sharper danger is a thinner entry route into exposed careers. The work that used to train juniors is often the first work a company can ask AI to compress.
I am not interested in the lazy version of this debate. “AI will create more jobs than it destroys” is not an answer. Neither is “AI replaces a task, therefore the profession is finished.” Software engineering, consulting, and general knowledge work have different demand curves, different ways of checking output, and different career ladders.