robotics is the next labour sink, and software is not going to hire the way it used to

What almost nobody says about the web and mobile waves is that we remember the platform and forget the labour. The boom is retold as Google and Amazon and the App Store, while the actual jobs went to the person who fixed a contact form for a dentist in Kolkata, the team that built a backend for a courier company in Coimbatore, and the tester with forty cheap Android phones that all behaved differently. India’s tech industry went from under a million people at the start of the century to about 5.95 million direct employees in FY26 by Nasscom’s estimate, which makes it the largest engineering labour sink this country has ever built. Most of it was unglamorous integration work that never makes it into a keynote. That sink is now close to full. Nasscom expects the sector to add about 135,000 net jobs this fiscal year, a 2.3% rise in headcount, while revenue grows 6.1% to $315 billion. Revenue is growing more than twice as fast as the people producing it, and that gap is the whole story. A week ago LTM’s chief delivery officer said his company is moving away from traditional entry-level engineering roles because AI has reshaped how it delivers.

Software used to soak up people because typing it was slow. Every page, every integration and every screen had to be written by a human, so every business that wanted software had to rent a lot of humans. That cost is collapsing. One engineer with good tools now does what used to take a small team, and the work that used to be handed to juniors, the glue code and the CRUD screens and the port to another framework, is exactly the work the tools do best.

We will write more software than ever, but jobs follow labour hours and not output, and the labour hours per feature are falling fast. Demand for software may keep rising. I just do not think it rises fast enough to rebuild the headcount, and the engineers who would have filled those seats will have to point their skills somewhere else.

I think robotics is the next one, and I mean labour sink in the way a heat sink works, as the place the excess energy goes when the last wave cools and a lot of engineers who learned to ship have to turn their skills towards something physical.

Software ate the businesses that could be turned into screens and mobile ate the ones that could be turned into a phone. What is left is the world of things that have to be picked up and moved and welded and sprayed and inspected, which by weight is most of the economy. It has sat outside the boom the whole time because a robot used to cost more than the person it replaced and needed a PhD to keep running.

The numbers say the gap is real. The world installed more than 600,000 industrial robots in 2025 by the International Federation of Robotics’ count. China took 354,000 of them and India took almost 10,500, which is up 15% on the year and about 27% a year since 2020, but still less than a thirtieth of China’s total. India’s robot density was around 26 per 10,000 manufacturing workers in the 2023 count, against a global average of 162 that has since climbed to 177. That is less of a mystery than a backlog.

It started changing in the last five years for reasons mostly borrowed from the phone in your pocket.

  1. The sensor stack got cheap. A phone camera, an inertial measurement unit, a GPS chip and a decent processor now cost less than dinner for four, and the same parts go into a robot that would have needed a defence budget in 2005.
  2. The software stack got free. ROS, simulation tools and open vision models mean a small team can get to a working prototype without inventing a middleware layer first.
  3. The models learned to see. The vision language models of the last two years can cut the integration time on a picking task from months of custom code to something closer to a week of examples, which is the kind of change that turns a research project into a purchase order. It is the same family of models that is shrinking software teams, which is why the two stories are really one.
  4. Labour got expensive and scarce. I will leave that argument for its own post, but the arithmetic under every deployment is that a pair of hands costs more each year and a machine costs less.

A website reaches a million users with one engineer and a load balancer, while a robot reaches a second factory only after somebody drives there and fixes the lighting and the network and the voltage and the way the workers load the input tray. The marginal cost of software is zero and the marginal cost of a robot is a van, two technicians and a maintenance contract, so the jobs scale with the number of sites, and the number of sites in India is enormous.

The robot is the easy part of the business because the site is the product, and every deployment drags a small army behind it. That army is where the engineering jobs sit, in integration and field support and maintenance and safety and data collection and fleet operations, and in the person who answers the phone at two in the morning when an arm stops mid cycle in a plant that has to ship by six.

I expect more engineers to work on robots in the next ten years than worked on apps in the last ten, and most of them will never work at a robot company, because the work is at the customer’s site.

The work will also be dirty in a way software never was. The robot has to survive a Chennai summer, a voltage fluctuation and a floor that was poured by a contractor who left early, and it has to work next to a human who has been doing the same task for twenty years and has strong opinions about the right way to do it. The Indian verticals are already full of this. We have robots going into sewers so that sanitation workers do not have to, drones spraying paddy fields, warehouse automation behind quick commerce and inspection machines on railway tracks, and every one of those is a site with its own mess and its own maintenance contract.

The humanoid gets the demo reel, while the money that actually reached the most people came from an arm in a welding cell and services work that paid by the hour.

I have been promised robot servants since I was a child and I still fold my own clothes, so I am not going to pretend the timeline is short. The first jobs will go to people who can make a camera see in bad light, make a motor repeat a motion ten thousand times and get a safety inspector to sign the form. The humanoid walking into your kitchen is a 2035 story at the earliest 🤞.

This is also a bet about the decade and not a description of today’s job board. Robotics hiring right now is small and tilted towards senior people, so nobody should expect a flood of junior openings next quarter. The move starts with the engineers willing to go to the floor and learn the job there. There is a version of this where the labour sink becomes a trap. The desk part of robotics, the integration code and the configuration and the glue, is what the next generation of models will compress, the same way they are compressing web work now. What compresses much more slowly is the site, with its van and its voltage and its angry client. That is why I am bullish on the decade and unsure about the one after it.

A decade is enough to plan a career around, and the leaders of the next one will come from people who spent it on factory floors with grease on their shoes, the way the last lot came from a garage and a broken deploy at two in the morning. If you are a young engineer asking what to learn, I would learn ROS and a PLC and how to solder, then spend a week on a real floor where the wifi is bad and the client is angry, because that week will teach you more than another certificate.

If you are betting your career on robots, tell me which vertical you would pick first, I am @troysk704.