When a Working Day Costs Six Dollars

A complete, working software product went from an idea to a live service in one working day. The labour bill for that day was six dollars and twenty-five cents. The same scope, priced at human agency rates, is usually around fifteen hundred dollars. Nobody reading that number should be thinking about code. The interesting figure is the 240x gap between what intellectual labour used to cost and what it costs now.

Economics has seen deflation before. We have watched it in semiconductors, in solar panels, in storage and bandwidth. Each time, the story followed a pattern: prices fall, the technology spreads, and the productive base of the economy reorganises around whoever can now afford the thing. What makes this round different is that the falling price is not a physical good. It is the cost of thinking through a problem and producing a finished result.

We have been writing about this for a year

This portal called the direction early. In February we analysed the numbers behind the 2026 AI hyperdeflation, where a World Bank economist put the fall in training and inference costs at 40x to 100x in twelve months, faster than Moore's Law ever moved. In April, before the rally in AI assets got comfortable, we argued that the utopia predictions were wrong because deflation is not prosperity, it is disruption with winners and losers. And in September we followed the logic to its uncomfortable conclusion in the post-human economy: if intellectual labour becomes nearly free and finite human needs stay finite, the next demand engine may not be human at all.

This week's case study is not a theory. It is a data point that landed inside the curve we drew.

What actually happened, in economic terms

A single person wrote a specification. An agent executed it: the application, the database, the mail system, the API, the deployment, the search-engine plumbing. About nine working hours of machine time. One human day of decisions. One hundred and four instructions from that human, roughly eleven per hour.

Read that as an input-output statement. One unit of skilled human judgement, plus a very large amount of cheap machine execution, produced an output that previously required a small team and several weeks. The scarce input shrank to judgement. Everything else became a commodity with a price that rounds to zero.

That is the same structural change that turned computing from a department into a utility, and it is happening to software production on a much shorter clock.

What it does to an economy

Three effects are worth separating, because they arrive on different timelines.

Prices fall first, and that is deflationary. When the marginal cost of producing software collapses, the price of software collapses with it. Companies that sell software as a service have to explain to their boards why the cost floor under their margins just disappeared. Investors looking at the AI rally have started to ask exactly this question, and economists are warning that some of the current valuations are priced for a world where none of this happens (CNBC on the correction risk).

Wages adjust second, and unevenly. Not every job that touches software disappears. The ones that were mostly execution do. The person who wrote the specification still had work. The people who would have spent three weeks implementing it did not. Sceptics are right that the industry's own economics are stranger than the marketing suggests (the Guardian’s look at the cost of AI slop), but the direction of travel on execution-heavy work is not really in dispute.

Measurement breaks third, and this is the part policymakers are not ready for. Statistics offices measure output through prices and wages. A product that used to show up as fifteen hundred dollars of billed labour now shows up as six dollars of an API invoice. The productivity gain is real, enormous, and largely invisible to the instruments that govern interest rates and fiscal policy. The Federal Reserve has already flagged that the evidence of AI's macroeconomic effect is still concentrated in investment rather than in measurable output (Fed note on the AI buildout), and Stanford's AI Index has been documenting the same gap between adoption and measurable economic effect (Stanford HAI, AI Index 2026).

Put plainly: an economy can get dramatically richer in real capability while its headline numbers barely move. That is a recipe for policy that reacts late.

The demand problem underneath

Every deflation story eventually runs into the same wall. If the cost of producing falls to near zero, and the income of the people who used to produce it falls with it, who buys the output?

Our own answer in September was that agents become customers: machine demand for machine output, a loop that can grow without human wages. It is not a comforting answer, but it is the logical one. And it raises the question the whole debate avoids. If the value created shows up neither in wages nor in measured prices, who owns it, and who does the economy then answer to?

Open questions, and we would like your take

• If a working day of skilled production costs six dollars, what is a wage actually for?
• Does deflation of this kind reach consumers as lower prices, or stop as higher margins for whoever owns the agent?
• If the productivity gain does not show up in the official statistics, how will central banks ever see the next crisis coming?
• Should an economy taxed on labour income start taxing machine output instead, and who would accept that?
• When the cost of building anything falls to near zero, what stops us from producing more than anyone needs?

We do not pretend to have settled answers. We do think these are now practical questions rather than thought experiments, and the full technical breakdown behind the six-dollar day is documented here: one business day, one complete product.

Tell us where you would draw the line between a deflation that spreads prosperity and one that concentrates it.