The AI gain is real, but so is the bill

AI both takes work and creates it, and the article argues that the real question is where the value ends up. It leans on Jevons' observation from the 1800s: when steam engines became more efficient, coal use rose, because making something cheaper gives people many new reasons to use more of it. The same is happening with thinking work. In one industry, reviewing a tender used to take two working days and now takes fifteen minutes, yet the same number of people are employed, because tenders that were once too small or too costly to assess are now worth pursuing. Cheaper software likewise means more software, more complexity and greater demand for the people who can manage it, which is why developers are in higher demand even as the machine is said to take over their work. The new bottleneck sits outside production, in demand, contracts, staff, banks, rules and accounting, which remain human and full of friction.

Geography is pulled in two directions. Signe, an experienced software architect in Ry, worked a few hours a week for 14 companies over the past year and was physically present at only four. AI handles heavy development work, so a local junior developer can draw on expertise that once required expensive specialists, and a company in Skive can reach the same technical skills as one in Østerbro. But Skive firms must also compete with Copenhagen, Hamburg and Austin for the specialists' time. Signe's price is set by global demand, not the local market. Ordinary development becomes cheaper and more widely available, while the few specialists who solve the hardest problems become more sought after and better paid. Work spreads out, but a larger share of earnings concentrates in a few hands.

The bill falls on public finances. The Danish welfare state relies heavily on tax on wages. When AI does work that people were once paid for, part of the income shifts from employees to the owners of the technology, and returns from ownership are taxed differently from wages. Many AI systems used by Danish firms are also foreign-owned. A Danish company that automates its bookkeeping with American AI still gets the work done, but part of the money goes to a US tech company rather than a Danish employee paying tax in Denmark. Companies save money, workers get better tools and the economy grows, yet Denmark could get richer from AI without the treasury getting correspondingly richer.

The article concludes that companies should embrace AI or risk being overrun, and that the size of the bill depends on two questions: whether the gain in breadth outweighs the loss in depth, and whether a way is found to tax value once it no longer passes through a payslip. It is the third part of a series based on the essay "Hvor meget menneske vil du ha'?".