Lights Out on Offshoring

We Ran Out of People

Somewhere in the late nineties, the amount of software the world wanted got permanently ahead of the number of people in America who could write it.

That is the whole origin story. Not strategy, not some spreadsheet in a boardroom. Demand went vertical and the supply of developers did not. Every company suddenly needed software, there were not enough of us to go around, and the ones who existed got expensive fast.

So the work went looking for people.

India Answered

India picked up the phone. Other countries did too, but India is the one that built an industry out of it.

I want to be careful here, because the lazy version of this story treats offshore development as a discount bin. It was not. The education systems were legitimately good. The engineering programs turned out people who knew what they were doing. English was widely spoken in a technical context, which mattered enormously. And yes, the price was competitive, in some cases dramatically so.

That combination is not a loophole. That is a country correctly reading a global shortage and building capacity to meet it. It worked, and it kept working for twenty years, and it built something enormous. Nasscom put India’s tech industry at $315 billion for the fiscal year that just ended, $246 billion of it exports, employing 5.95 million people. That is not a side hustle. That is a pillar holding up a national economy.

But It Took a Week to Move an Inch

Here is the part that anyone who has worked with an offshore team knows in their bones.

The time zones did not line up. That was not a scheduling annoyance, it was the defining constraint on everything. And language was almost always a barrier somewhere, not because anybody lacked skill, but because being precise in a second language under deadline pressure is genuinely hard, and software is nothing but precision.

Put those two together and you get the loop.

Day one, you write up the task and hand it off. Day two, the work happens while you are asleep. Day three, you read what came back and write feedback. Day four, the rework happens while you are asleep again. If you were lucky, that was the end of it. If the spec had any ambiguity in it, and specs always have ambiguity in them, you went around the whole thing again.

A feature that should have taken an afternoon took a week. Not because anyone was slow. Every person in that loop was working hard and doing their job well. The loop was the problem, and the loop was made out of geography.

That was the deal. You got more hands at a better rate and you paid for it in latency. For a long time that trade made sense, because the alternative was not having the hands at all.

Every One of Those Problems Just Died

Then it went away. All of it, at once, and not gradually.

The agent works when I work. No handoff, no overnight, no waiting for a continent to wake up. It is two in the afternoon for me and it is two in the afternoon for the work.

The language molds to me. I do not write a specification and then hope it survives translation. I say the thing, and if it lands wrong I say it differently ten seconds later, and there is no cost to that. Nobody is embarrassed. Nobody has to go ask a manager whether it is okay to push back. The ambiguity that used to burn two days gets burned down in a sentence.

And the feedback loop is measured in minutes. Not days. Minutes. That one change is the entire ballgame, because the four-day loop was never really about cost. It was about how long it took to find out you were wrong.

Then There Is the Price, Which Is the Scary Part

For a few hundred dollars a month I can keep four or five agents working at the same time.

Not full time the way a person is full time. Full time the way a machine is full time. No nights, no weekends, no holidays, no PTO, no onboarding, no attrition, no notice period. Three in the morning on a Sunday is a working hour.

Put that against the fully loaded cost of one offshore developer, with the account management and the coordination overhead and the rework included, and it is not close. It is not the same category of number.

That comparison should make everybody uncomfortable, and it does make me uncomfortable. But pretending it is not true does not help anyone.

From the Assembly Line to Lights Out

Manufacturing already went through this, so we know the shape of it.

The assembly line was the first move. It did not get rid of the worker, it reorganized the worker, and output exploded. That is roughly what offshore development was for software. Take the work, break it into assignable pieces, route it to where the labor is, get more out per dollar in.

Then came lights-out manufacturing. Fully automated production running with nobody on the floor. The name comes from the fact that machines do not need to see what they are doing. FANUC has run a plant in Japan this way since 2001, robots building robots, around fifty a shift, unsupervised for up to thirty days at a stretch. My favorite detail is that they turn the heat and the air conditioning off too, because there is nobody in there to be cold.

That is the move we are in the middle of right now. Not a better assembly line. No assembly line.

They Are Already Saying It Out Loud

This is not a prediction I am making from the cheap seats. The companies whose entire business is offshore delivery are describing it themselves, on earnings calls, to their own investors.

Across a full year, TCS, Infosys, Wipro, and HCL added 3,910 people between them. Four of the largest technology employers on the planet, combined, added fewer people than a mid-sized company. TCS dropped eleven thousand in a single quarter. All four told investors they are leaning on more AI to deliver work for clients.

Three months later, HCL’s CEO gave the thing a name. He called it AI deflation and guided revenue down three to five percent, with more possibly behind it. TCS revenue slipped year over year and its CEO called that degrowth, which is a hell of a word to have to say into a microphone.

In fairness, HCL grew eleven percent last year. It is not a uniform collapse and I am not going to pretend it is. But look at what that growth is attached to. Revenue climbs while headcount sits flat. The link between people and money, which is the entire mechanical basis of selling development by the seat, is coming apart. When your business is billing for hours and your hours stop scaling with your revenue, you do not have a growth story. You have a countdown.

I Am Not Cheering About This

I want to stop and say this plainly, because there is a version of this post that reads as America winning something. It isn’t that. We are all both gaining and losing at the same time.

Five plus million people. That is who works in that industry. Behind that number are careers people built on purpose, degrees people took on debt to earn, families that reached the middle class on the strength of this work, and an export economy a country organized itself around for two decades. If this goes the way I think it goes, that is a real hit landing on real people whose only mistake was being very good at the exact thing the world asked them to be good at.

I do hope this pulls talent back into the United States, and I hope it makes what we build here stronger. I am not going to be coy about hoping that. But wanting that is not the same as wanting the other thing, and I am not going to dress up somebody else’s economic damage as a win for my side. Both of those are true at the same time and they do not cancel each other out. I am also certain that the issues abroad are ramping up here next.

So What Is the Service?

Here is the question I cannot get around.

I was always the one writing the specification. That has not changed. I still sit down and describe what needs to exist, the same way I did when the description was headed for a team eight thousand miles away. The specification was the actual thing I was producing. It always was.

What changed is what happens to it after I hand it over. It used to go into a four-day loop and come back partly right. Now it goes into a loop measured in minutes and comes back partly right a great deal faster, I can run five of those at once, and it costs me almost nothing.

So what am I buying from a services firm? Not capacity, I have capacity. Not savings, I am already cheaper. Not access to skills I do not have, because the model has more breadth than any team I could realistically staff. What was being sold was people at a price, and the price of people just stopped being the thing standing between me and working software.

Turn Off the Heat Too

The lights are going out on offshoring as we knew it.

Not on India, and not on the talent, which was always real and is not going anywhere. On the specific arrangement where you buy engineering hours by the seat because the hours are cheaper somewhere else. That model was an answer to a shortage, and the shortage is over. Something will take its place, and I do not know what it looks like yet. Real domain depth, probably. Verification and accountability, which are going to matter far more than anybody is currently admitting. Owning an outcome instead of renting out hands. Whoever is working that out right now is who is still here in five years.

But whatever it turns into, it will not be the thing that has been sold for the last twenty years. That thing is finished, and the people who built it are the ones telling us so.

FANUC turns off the heat because there is nobody in the building to be cold. I keep coming back to that. We are building rooms that nobody has to stand in, and we have not spent nearly enough time thinking about where everybody who used to stand in them is supposed to go.