Who Will Buy What Your AI Builds?

Every company wants to cut costs. But if every company cuts the incomes of its customers, who will be left to buy?

Most AI pitch decks carry the same slide. It shows the hours of human work the product removes. Investors love that slide, and I understand why. For one company, cost saved is value created.

But hold all the decks together. Each one removes a salary somewhere. A salary is not only a cost. It is also someone's spending. The analyst you replaced was buying software. She paid school fees. She kept three subscriptions; maybe one of them was yours.

An economy is not a warehouse. It is a flow. Money must pass through people before it returns to you. Earlier this year, two economists modelled exactly this. Brett Hemenway Falk and Gerry Tsoukalas call it the AI layoff trap. Each firm that automates keeps the whole saving. The lost spending falls on everyone, rivals included. So every firm races to automate, eroding the demand it depends on. At the limit, they write, firms reach "boundless productivity and zero demand".

Not every economist agrees. Past technologies lowered prices and raised incomes, and demand adjusted. AI may too. But no founder should bet the market on a "may".

The Moving Moat

When anyone can generate the output, the output itself stops being scarce. Stanford's 2025 AI Index tracked what a GPT-3.5-level model costs to run. That price fell more than 280-fold between late 2022 and late 2024.

So value does not vanish. It moves into trust, distribution and taste. A thousand AI-written reports now cost almost nothing. The one report a board trusts is worth a great deal. Ask what your customers will still find scarce in three years. Build there.

The Talent Gap

Senior judgement grows out of junior work. People become editors by writing weak drafts first.

This is no longer a guess. Stanford economists Erik Brynjolfsson, Bharat Chandar and Ruyu Chen track American payroll data. Their August 2026 update looks at young workers, aged 22 to 25, in the most AI-exposed jobs. Their employment is now about 19 per cent lower than if it had kept pace with young peers in less-exposed jobs. Experienced workers show no such gap. The drop comes mostly from fewer hires.

So where will tomorrow's experienced people come from? Aviation met this question early. In 2013 and 2017, America's aviation regulator issued safety alerts. They urged airlines to let pilots fly manually more often. Autopilot is not bad. Overreliance on it can erode the manual skills pilots need when automation fails. Protect some struggle on purpose. Let junior employees do work that AI could finish faster. Book it as training, not waste.

Build for Circulation

Some AI companies sit in the middle of a process and charge for passage. Others make many people more capable and share in what they create. Only the second kind builds its own market.

Take Shopify. A Deloitte study it commissioned used 2021 data. Merchants earned about 38 dollars for every dollar Shopify earned. They supported some five million jobs. The figures come from Shopify, so hold them loosely. But the direction is clear. The tool created sellers. The sellers then spent on the tool.

Run the same test on your product. Does it make 10,000 people more productive? Then you have 10,000 customers who can grow. Does it make them unnecessary? Then you have one customer. And that customer will squeeze your margin next.

I am not arguing against AI. I teach with it. I write with it. The gains are real. I am asking founders to look one step past the demo. A market is people with the means to buy. Abundance that never passes through people is not a market. It is inventory.

So, before the next pitch, ask: who will buy what your AI builds? Make sure the honest answer is not "fewer people every year".

(The author is an academic researcher based in Jaipur. Views personal.)