The Conversation: Kartik Rao, Vahan AI

Every signal this week has been about white-collar AI exposure - programmers, analysts, knowledge workers. Kartik Rao works at the other end of India's labour market entirely. Vahan AI has placed over one million blue and grey collar workers in jobs across 920 cities, using conversational AI to reach people that no recruiter, job board, or government portal has ever reliably reached. This conversation is about what AI actually looks like when the workforce you are serving delivers your groceries, stitches your clothes, and builds your city.

Q: All this week's signals are about white-collar AI exposure. Programmers at 74%, analysts at 64%, knowledge workers across the board. The blue and grey collar workforce barely appears in any of this data. Is it actually protected from AI disruption - or has nobody built the measurement framework yet?

AI has already reached the blue collar surface. We have partnered with a company that makes a physical product workers place on their head that records productivity in real time. That is AI operating in a physical, intimate space - not measuring knowledge work but measuring the pace and quality of physical work. The touch points are just different. For a blue collar worker, AI might arrive as a voice monitor, or as a WhatsApp chatbot that speaks in their dialect, or as a data collection tool that trains models on how they actually move and work. White collar AI models have matured because there was training data. Blue collar models are immature because nobody has collected that data yet at scale. That is changing. I know companies in India and globally that have raised significant capital on exactly one proposition: we have data on these workers, and we will find a way to monetise it. The measurement framework is being built right now. The exposure numbers will follow.

Q: Vahan uses AI to automate the hiring of blue collar workers. Those same workers are now competing for jobs in quick commerce, logistics, and manufacturing - sectors where AI and automation are also changing the work itself. Are you simultaneously providing a solution and part of the problem?

I do not think AI is yet changing pure-play blue collar physical work in any fundamental way. We are still far from the day a drone delivers milk to your door or a machine does embroidery with the skill of a craftsperson. Today cars are manufactured without human intervention - but that took decades and enormous capital. In logistics and quick commerce, supply is still the problem. Our clients come back saying they need more blue collar workers, not fewer. There is demand for a lakh of people a month and we are placing thirty to forty thousand. If AI was truly displacing physical workers at scale, that gap would not exist. What AI is doing in this sector right now is monitoring, measuring, and enhancing - not replacing. The ratio of what AI can actually do in physical work is about two out of ten, not seven out of ten. The interesting observation I will add is this: it is the manager getting affected, not the executor. AI is replacing the recruiter who makes the qualifying calls. The worker being recruited is untouched.

Q: Vahan's human-in-the-loop model has AI handling 80 to 90 percent of the top of the funnel. What does the recruiter role actually look like today versus two years ago - and what does it look like two years from now?

It has moved from qualification to conversion. Two years ago, a recruiter spent most of their time calling people, asking basic questions, filtering out unqualified candidates. Today AI does all of that. The recruiter enters the conversation when it matters - when a candidate is qualified and needs to be converted into a placed worker. Productivity has increased significantly. The question you are really asking is whether conversion itself goes to a bot. Honestly, in a year it might. The stakes are high enough that right now you would not trust AI with it - but AI is already capable of conducting interviews. There is a great analogy I keep thinking about. Planes can fly themselves. Someone once said that in future you will not need two pilots - you will need a pilot and a dog. The pilot's job is to feed the dog and the dog's job is to make sure the pilot does not touch anything. But nobody boards a plane today with no pilot at all - because human confidence in the system has not caught up with technical capability. That is exactly where blue collar AI recruiting is. The capability exists. The trust is being built.

Q: India has hundreds of languages and dialects. The workers hardest to reach - the ones in the most precarious employment - are also the ones whose language and context AI handles worst. What keeps you up at night about who you are not reaching?

Honestly, language has moved down the worry list. AI is solving that faster than anyone expected. Every month it becomes less of a problem. What keeps me up now is the quality of engagement - not whether we can reach someone, but whether the interaction we have with them when we do reach them is actually good. What is the latency in response? Does the candidate feel heard? Is there something qualitative and human in the exchange that creates trust and converts? We have moved from transactional problems to qualitative ones. That shift itself tells you how fast the technology has moved. A small thing that illustrates this - I barely type anymore. I just hold the shift button, speak my jumbled thoughts, and AI cleans it up and shapes it. That is not a white collar AI story. That is what happens when language barriers start to dissolve at scale.

Q: What are you getting wrong? What assumption are you carrying into this year that might not survive contact with reality?

Two things, and I think both apply well beyond Vahan. First: people are overestimating what AI can do and underestimating the cost of finding out. I see companies firing ten thousand engineers only to discover a week later that everything broke and they did not have the human context to fix it. We have the same disease at a smaller scale - there are places we should have used AI more and have not, because we were cautious, and places we have over-indexed on it and paid the price. The calibration is genuinely hard. Second - and this is the one I feel most strongly about - people are using AI to be dumber instead of smarter. Instead of using it as a thought partner or a critic, people are using it as an intern to write for them. AI will always give you something. And because it sounds like you, you never fully critique it. You look at the output and think it is an extension of yourself, so you approve it without the scrutiny you would apply to someone else's work. Ninety-nine percent of people are doing this. They are taking the path of least resistance. The people who will benefit most from AI are the ones who use it to sharpen their thinking - not replace it.

Kartik Rao is a business leader at Vahan AI, the Bengaluru-based platform backed by Khosla Ventures, Y Combinator, and Airtel that has placed over one million blue and grey collar workers in jobs across India. Vahan is expanding into textile and electronics manufacturing as India pushes toward a trillion-dollar manufacturing output.

The question Kartik answered that most white-collar AI commentary avoids: not what AI will do to workers, but what workers do to AI when the tools arrive in their hands. The answer is the same whether you are a knowledge worker in Bengaluru or a delivery rider in Patna - you use it the way that feels easiest, which is not always the way that makes you better.