Services vs. Product in the Era of AI
AI broke the linear math of services and supercharged the compounding math of product. But as models commoditize depth, the scarce skill becomes orchestration, and nobody has practiced it longer than services people.
For thirty years, the software industry ran on two engines. One sold time: the services firm, billing by the hour, scaling by headcount, perfected by Accenture, Infosys, TCS, and a thousand agencies from Lahore to Lviv. The other sold artifacts: the product company that built a thing once and sold it a million times. Both got rich, but they were never the same business. AI has just made the difference brutal.
The hit
A services company's revenue is a linear equation: people x hours x rate. Everything in the model, from the pyramid of many juniors under a few seniors to the labor arbitrage between geographies, assumes that software is made of human hours and that hours are scarce.
AI attacked the equation at its root. Code generation collapsed the junior layer, the base of the pyramid that made the margins work. Clients who once accepted a forty-person team now ask why twelve people and AI can't do it, at a fraction of the price. Worse, every productivity gain a services firm adopts cuts into its own billable hours. It is a strange business when your best new tool shrinks the thing you sell. The market noticed. In 2025 TCS carried out the largest layoffs in its history, and Accenture told investors it would part ways with staff who could not be reskilled for AI. Time, as a product, is deflating.
The win
Product economics run the other way. Revenue is decoupled from headcount, marginal cost trends toward zero, and every improvement compounds into margin instead of eating it. For a product company, AI is pure leverage. The same team ships faster, the software gets smarter, and the smarter version commands a higher price. The defining valuations of this decade are all product: Nvidia past four trillion dollars, the model labs, the entire infrastructure layer.
To be fair, not every product won. Thin wrappers over someone else's model got commoditized as fast as junior hours did. The win went to depth: proprietary data, distribution, a problem owned for years.
"One GPU company is now worth roughly three years of the entire world's IT services billings. The market priced the hour and bought the artifact."
Two kinds of mind
Services breed a horizontal mind. You parachute into any domain (banking on Monday, logistics in March, a government portal by Q3), extract requirements from people who cannot articulate them, ship against a deadline that was political before it was technical, and move on. The craft is speed, context switching, and figuring out what an organization actually needs versus what it asked for. Done means delivered and invoiced.
Product breeds a vertical mind. You live inside one problem for years. The craft is saying no, sequencing, taste, and treating technical debt as a strategic variable rather than someone else's inheritance. Done means adopted, retained, and measured, which in practice means never.
Crossing over is disorienting. The services engineer entering product feels slow, suffocated by a single domain. The product engineer thrown into services feels shallow and cut off from outcomes. Each quietly suspects the other isn't doing real engineering.
The services edge
There is a twist, though. When AI knows everything, when vertical knowledge is one prompt away, depth itself gets partially commoditized. What AI does not yet supply is the thing services veterans were forged in: orchestration. Decomposing a vague mandate into work. Running a team, except the juniors are now agents. Knowing which of seven plausible solutions will survive contact with a real client, a real regulator, a real legacy system. The pyramid is dead as a business model, but it is exactly the shape of AI-era work: one experienced human directing many agents. Services people have rehearsed this their entire careers.
So the likely winners, at least among individuals, are hybrids. People who keep the breadth and orchestration instincts of services, pick up the ownership and taste of product, and let AI do the heavy lifting.
A world at war
None of this is happening in a stable world, and the instability is part of the mechanism. The services model was a child of globalization: open borders for contracts, geography as just an arbitrage variable. The wars of this decade broke that assumption. Russia's invasion of Ukraine wiped out outsourcing corridors overnight, and sanctions, data sovereignty laws, and friend-shoring have put borders back into an industry that pretended they didn't exist.
The wars also proved that software is now strategic materiel. Cheap computer vision drones reshaped the Ukrainian front. The genocide in Gaza surfaced contested reports of algorithmic targeting and dragged AI ethics out of conference rooms and into the streets. Gulf states started converting oil wealth into compute wealth, betting that the next barrel of strategic power is a megawatt feeding a GPU. Defense product firms like Palantir and Anduril went from curiosities to primes. Every one of these shifts rewards owning the artifact. None of them rewards renting out hours across a border that might close.
For services exporting economies like Pakistan, India, and Eastern Europe, the reading cuts both ways. It is uncomfortable, because the export model that built our industries is deflating just as the world fragments. It is also hopeful, because AI has collapsed the capital cost of building product. A small team in Lahore can now ship a deep vertical product that would have needed a Valley-sized budget in 2019. Geography decided who could do services. It no longer decides who can do product.
Who wins
Among nations, the ones with energy, compute, and the trust to host other people's data. Among companies, deep products with proprietary data and distribution, plus the services firms that manage to reinvent themselves as outcome owners, selling deployed results instead of hours. And among individuals, the orchestrators: people with the horizontal reflexes of services, the vertical patience of product, and the humility to let the machine do the typing while they keep custody of the judgment.
The billable hour is dying. Judgment isn't. This era belongs to the people who can tell the difference.