Chapter 22
Mortgages Priced on Salaries
The Utopia of the Agentic Enterprise, Part IV. Doing Less
Income from cognitive labour supports more than the people who earn it. Banks price mortgages against it, and the taxes on it fund government budgets. A business case for stopping a piece of work or handing it to an agent stops at the edge of the company. The saving it books is an income someone stops receiving, somewhere the ledger doesn’t reach, and the list of who consumes the work and who makes up for it stops at the same edge. Some of that income was a supplier’s, paid to people who know the client’s workarounds better than the client does.
None of this is a forecast. Timelines and bargaining power will vary by place and sector, and a market can reprice a profession without announcing the new price. Think of it as a map of whose income the saving was.
Business-process salaries in India grew fast for a generation. India’s IT and business-process sector earns well over $200 billion a year, most of it from exports.1 The Philippine outsourcing industry earned about $38 billion in 2024 and employs around 1.8 million people.2 In both countries the chain from an individual salary to the national accounts runs through leverage. The bank lent against the salary and the government budgeted against the tax on it, both on the assumption that knowledge-work exports would keep growing.
So a saving booked in one country can be income that stops in another. It wouldn’t stop overnight. The salary stagnates while the mortgage payment stays where it was, which is why the correction would register as a financial adjustment before anyone read it as a labour-market one.
The same chain runs, shorter, through the tax. Personal income tax and social security contributions make up about half of all tax revenue across the OECD.3 When knowledge-work income reprices, that revenue would shrink before anything replaces it, because the new value concentrates in fewer companies, and in jurisdictions set up to attract it.
None of these costs is in the business case that produced them. The person with the mortgage and the treasury with the budget pay them, and neither was in the meeting that approved the project. It’s the arrangement that makes a customer fight the support bot, so that the institution saves its own time by spending theirs. Here the company spends someone else’s income instead, at the scale of a national economy, and none of the people it costs are on its books.
Offshore providers that have run a client’s processes for years know which data fields are unreliable and which business rules contradict each other. Much of it is knowledge of the client’s own workarounds: the reconciliation that exists because two of the client’s systems disagree, the manual step inserted after an old audit finding that nobody at the client can explain any more. They have been the client’s duct tapers, and they know where all the tape is.
One argument for insourcing goes like this. When AI cuts the cost of the engineering itself, what remains is coordination: translating goals across distance and time zones, and managing handoffs between organisations, which, as anyone on a status call at an hour that suits neither office knows, is the expensive part. A small AI-augmented team close to the operation can now match an offshore team on total engagement cost, because the wage gap that justified the distance has narrowed while the coordination overhead hasn’t. The argument is sound, as far as it goes. But it is still a decision someone makes, and it ends a relationship with people who hold years of knowledge about the business, including the parts its own staff never wrote down. Some of that knowledge leaves with them.
Of course, the offshore companies are not standing still. Infosys launched a suite of services in 2024 that tracks the compliance of a client’s AI projects and screens what goes into and comes out of the models,4 which is roughly the duct taper offering to inspect the tape. But clients who have spent years buying offshore providers by the head don’t automatically see them as anything else. And their pyramids rest, like a consultancy’s, on graduates who learn on routine work. India’s IT sector hired between 60,000 and 70,000 graduates in the year to March 2024, the fewest in 20 years and about a third of its yearly intake before the pandemic. The reasons given were weak demand and a surplus of earlier recruits, and the companies planned to hire more the following year.5 But the routine coding and testing that trained those graduates is the work the models now do, and whoever audits the models’ output has to have learned the work somewhere.
Diversified economies spread the repricing across industries. Economies concentrated in knowledge-work exports meet it directly, and the mortgages and budgets resting on those exports amplify it. Either way the cost travels in the same direction: outward from the company that booked the saving, and downward to whoever’s income the saving was.
A company can’t stop that on its own, but it can see its own part. A business case can give each saving a row for whose income it was. And before a supplier contract ends, the client can pay for what the supplier knows about the client’s workarounds to be written down. An agent can draft much of that from the supplier’s tickets and runbooks, and the supplier’s staff can correct it while they’re still being paid to.
Checklist
- What economy, sector and city does your organisation depend on, and what does that system assume about knowledge-work income?
- Where are financial commitments such as real estate, credit and public budgets leveraged against that income continuing to grow?
- Which of these forces can you influence, and which can you only position against?
- Before a supplier contract ends, who writes down what the supplier knows about your workarounds, and are its staff still paid to correct it?