AI Has Repriced Outsourcing. Most Founders Are Still Paying the Old Price.

There is a quiet repricing happening in global software outsourcing, and most of the people paying the invoices have not noticed yet.

The shift is simple to state. For twenty years, outsourcing has been sold by the hour and priced on headcount, and a large share of those hours went on work that is repetitive and well specified. That is exactly the category of work AI now compresses from hours into minutes. The cost of producing outsourced software has fallen. The price most buyers pay has not. The gap between those two lines is the story, and it will define who wins and loses in outsourcing over the next decade.

If you buy outsourced development, that gap is landing directly on your invoice, whether you have noticed or not.

The Pressure Is Already Visible

You do not have to take the argument on faith. It is showing up in public numbers.

Infosys, India's second largest IT services firm, reported a quarter in which revenue grew fourteen percent year on year to roughly five billion dollars, yet still missed analyst estimates, and the company trimmed its annual growth forecast, with reports pointing to clients holding back spending as AI disrupts traditional software services [1].

The most telling number in that release was not the miss. It was the mix: AI services made up 8.2 percent of Infosys revenue, up from 5.5 percent a single quarter earlier [1]. The traditional work is softening while the AI-related work grows fast. The giants of Indian IT are not being destroyed by AI. They are being forced to migrate off the model that built them, in public, one quarter at a time.

That model is the one most founders still buy: teams billed by the hour, priced on headcount, doing work that is largely repetitive and well specified.

The Research Behind the Trend

Goldman Sachs put numbers on exactly this seam in a research report on generative AI and the Indian economy, "Gen-AI in India: Jobs Crosswinds, Productivity Tailwinds" [2].

Goldman estimates that generative AI could perform between 9 and 17 percent of the tasks currently done by India's non-agricultural workforce, depending on how capable the technology becomes [2]. But the distribution of that risk is the real finding. Only 8 to 12 percent of non-agricultural employment faces substitution risk, while 42 to 48 percent of workers are likely to see their roles complemented by AI, freeing them for higher-value work [2].

And where does the substitution risk concentrate? In the report's words, routine business functions in IT-enabled business process services face relatively higher substitution risk because they involve repetitive and codifiable tasks [2].

Repetitive and codifiable tasks. That is a precise description of the work the outsourcing industry has sold by the hour for two decades. Boilerplate code. Routine integrations. Manual testing. Documentation. Status reports about all of the above.

Goldman also projects that widespread AI adoption could add around 0.4 percentage points to India's annual labour productivity growth over the coming decade, with a range from 0.1 to as much as 0.8 depending on how capable the technology becomes [2]. The productivity is going up. The open question is who captures it: the vendor or you.

Why This Lands on Your Invoice

Here is the uncomfortable arithmetic for anyone paying for outsourced development by the hour.

A meaningful share of the billable work in a traditional engagement is exactly the repetitive, codifiable kind that AI now compresses dramatically. If your rate card looks the same as it did a few years ago, one of two things is happening. Either your vendor is not using AI, and you are paying for slow work. Or your vendor is using AI, and you are paying human prices for machine-speed work while the margin quietly moves to their side of the table.

Neither of those is a scandal. It is simply what happens in every industry when the cost of production drops faster than the price of the product. The customers who notice first get the benefit. The customers who notice last fund the transition.

Clients holding back spending on traditional IT services, as the reporting around the large vendors describes [1], suggests plenty of buyers have started to notice.

The Talent Was Never the Problem

It would be easy to read all this as "outsourcing to India is over". I think that reading is exactly backwards, and I say that as someone who has built products with teams in India for years.

The talent was never the weak point of the model. The pricing was. India produces some of the strongest engineers I have worked with anywhere, and the best of them are not fighting AI. They are compounding with it. A small senior team working AI-natively, with agents handling the repetitive layer and humans owning architecture, judgment, and quality, now ships what a much larger team used to.

The research backs the talent story: nearly half the workforce complemented, a fraction substituted [2]. The engineers whose work was mostly judgment were never the ones at risk. The hours were at risk. The people were not.

Which means something genuinely exciting for anyone building. Some of the strongest engineering value in global tech is a senior Indian team with AI in its hands, bought on outcomes rather than hours. The cost of great engineering has quietly fallen for buyers who structure the deal correctly, while most buyers remain structured incorrectly.

That mispricing is a window. It will not stay open forever, because markets eventually reprice. But structural transitions in how services are bought take years, not months, and the advantage belongs to whoever moves early within them.

How to Buy Outsourcing in the AI Era

If you are engaging or re-negotiating an outsourced team, the fix is less about squeezing rates and more about changing what you buy. In practice, this is what I look for.

1. Buy outcomes, not hours. Define the deliverable, the quality bar, and the acceptance tests. Let the team decide how much of the path is AI-assisted, because the good ones will use it aggressively and you want them to.
2. Ask how the team uses AI, specifically. Which tools, on which parts of the work, with what review process. A vendor who cannot answer in detail is either behind or hiding the answer.
3. Insist on human review where it matters. AI-accelerated does not mean unreviewed. Security, data handling, and architecture decisions need senior human eyes regardless of how the code was produced.
4. Keep the team small and senior. The old model priced large teams of mixed experience. The new model rewards a handful of experienced engineers amplified by AI. Fewer people, better people, faster output.
5. Measure velocity against outcomes, not activity. Tickets closed and hours logged tell you almost nothing now. Working software in production tells you everything.

Where This Goes

Put the pieces together and the picture is coherent. The traditional hourly model is under visible pressure, which is what the results of the large vendors show [1]. The underlying talent, amplified by AI, is becoming more productive, which is what the research projects [2]. The value has not left Indian engineering. It has moved from the hours to the outcomes, and pricing has not fully caught up.

For founders, that is not a threat to manage. It is one of the better arbitrage opportunities in tech, available to anyone willing to change how they buy. And it rewards the same skill every structural shift rewards: noticing early that the thing you are paying for is no longer the thing being made.

I have spent years building with outsourced and offshore teams, including a team in India I trust with my own projects. If you are exploring this and want an honest steer, or an introduction to a team I actually ship with, see how I help founders outsource software development, or get in touch. I am happy to share what has worked, what has burned me, and what I would do differently starting today.