When an interviewer hailed InMobi as India's first unicorn, one of its co-founders offered a history lesson. "Way back," he began, "India was a land of innovators, discoverers, inventors. We innovated zero, Ayurveda. These are all global products from India." InMobi, he added, with what seemed like feigned modesty, was "resurrecting that magic India once had for entrepreneurship."

The supposed resurrection is Glance, an AI shopping app. “See a new version of you,” its website coos. All you have to do is upload a selfie, and the app obligingly assembles an avatar dressed in clothes pulled from various online marketplaces.

I took a bad selfie on Glance. My avatar, unimpeachably styled, bore a passing resemblance to me. She always faced the camera with unsettling composure and seemed to possess an endless wardrobe. We both looked best in terracotta. (Glance called it our “standout shade,” insisting it highlighted “our warm, deep undertones.”) It doesn’t help that she wears my face with more aplomb than I can muster. At one point, Glance hallucinated a perfect black polka-dot dress, the only thing I wanted to buy. There was another problem. In one of its first “looks,” it put my head on backwards in a yoga pose.

Long before AI, India's unicorns had already discovered that the most profitable position in any market is the one just above the people doing the actual work. So, a generation of middlemen had come of age, bankrolled by venture capital that had every incentive to keep it that way. "India's most celebrated start-ups – its unicorns – have very little innovation to brag about," writes Outlook Business correspondent Nabodita Ganguly. Many of them are, in effect, professional hangers-on, extracting value from infrastructure they did not build.

In India’s caste-inflected economy, the middleman casts a long shadow. Traipsing between laborers and the market, he collects his commission while leaving the grunt work of production and sale to others. In fiction, he is a stock villain, marked by a miserly relish for extracting credit from the most wretched.

Now the middleman has migrated online, most visibly in the form of the platform economy, a virtual marketplace where service providers meet consumers. As Nick Srnicek argues in Platform Capitalism, platforms "provide the basic infrastructure to mediate between different groups" without needing to "build a marketplace from the ground up," relying instead on "pre-existing infrastructure" and "cheap marginal costs." In India, digital intermediaries like Zomato and Swiggy have built vast enterprises on the backs of delivery workers, skimming a commission off every order. Zepto, a prominent player in this space, was founded by two Stanford dropouts who returned to address India’s “grocery problem.” Here, at last, were India's boy geniuses, its Zuckerbergs. What they built, however, was a network of dark stores and delivery riders, a marginally faster version of the neighbourhood shop that had always existed. The novelty was not in the product but in how it was made to appear so.

One might attribute the underwhelming "innovations" to a prolonged history of keeping engineering and manual labour apart. As Ajantha Subramanian argues in The Caste of Merit, technical knowledge in India was gradually lifted out of the hands of lower-caste artisans and refashioned into white-collar professions for privileged castes. When engineering was "purified" by severing it from the supposedly "tainted" labour of marginalised castes, the result was a profession that exalted the conceptual over the manual.

Not to mention, as Ganguly observes, "...VCs in India are designed for speed, predictable revenue curves, and returns that can be quickly explained to limited partners." This preference for rapid, legible returns follows the older economic logic of the middleman straddling between labour and capital. In his 1907 book Indian Jottings, Edward F. Elwin wrote of middlemen who bought brassware from "industrious" city workers at a "ruinous loss." The same merchants, he noted, were also the district's chief moneylenders, advancing credit at "an appalling rate of interest."

With merit increasingly narrowed to "raw intelligence" and "analytical competence" – traits more easily channelled into software and management than into the construction of physical infrastructure – engineers gravitated toward the role of the middleman.

Once a "much-maligned economic actor" of India's agricultural economy, the middleman finds a second life in the country's AI start-ups, their digital descendants. With platform capitalism beginning to show its ceiling, and more people disillusioned by its gratuitous valuations and its claims to social utility, tech founders have had to find a new way to insert themselves into the supply chain. Now, with AI, they – modern-day “AI entrepreneurs,” if you will – can engineer add-ons, dress them up as indispensable services, and sell the story of innovation without the inconvenience of actually having to innovate.

At India's AI Impact Summit in New Delhi this February, the nation's technological ambitions were on full display, along with a Chinese-made robot, presented by Galgotias University as a "domestic" innovation. Soon after, there were memes aplenty, but as Manisha Pande writes of the hoopla: "We are very good at enjoying the spectacle of humiliation. We are less good at locating responsibility."

With innovation becoming a nationalist obligation, the temptation to fake it is never far behind. "Technocrats and politicians envisioned the entrepreneur as the 'global Indian' – the highly educated, wealthy visionary who could innovate industries and create value for India," writes Lilly Irani in Chasing Innovation. These figures, she notes, are trained to "look for opportunities to take on projects and redirect their lives to add value."

For entrepreneurs, a colossal population means a colossal number of prospective consumers. With AI services thrown into the mix, the economics look even more favourable, because cheaper capital yields greater returns. The trickiest part is securing the buy-in needed to convince such a large population of the AI middleman's necessity.

The sell, when it works, follows a reliable script. Take, for instance, what was proposed to solve what digital rights lawyer Apar Gupta describes as a "lack of labour-intensive… [and] even professional lower blue-collar jobs in India": an AI-powered jobs platform. Such reasoning frames unemployment as a mere “information asymmetry” addressable by an app, letting us off the hook in the process. In AI Snake Oil, Arvind Narayanan and Sayash Kapoor argue that the appetite for add-on services often betrays institutional failure. Broken institutions, unwilling to confess to structural decay, turn to AI as a talisman, mistaking the technology for a silver bullet.

There is an almost "evangelical zeal" in treating AI as a "panacea" for deep-seated social problems, affirms Gupta. This is how the story of middlemen goes. At first glance, the promise of yet another app creates the impression that something is being done. The middlemen make it appear that redress is within reach – or possibly already underway – while in reality it drifts further away.

The farce of the middleman economy, coupled with a risk-averse government, nudges citizens to redirect their dissatisfaction away from structures of power and toward the so-called "failures of imagination" when projects fail to materialise.

Such practices, Irani argues, "bend away from the slow, threatening work of building social movements." When entrepreneurs present systemic issues as isolated targets for intervention, the imperative for large-scale mobilisation around grounded welfare projects is eclipsed. The effect is a "prophylaxis against protest," leaving structural inequities intact as problems indefinitely await redress.

Stanford’s Global AI Vibrancy Tool, a data-driven dashboard that ranks countries by AI strength, placed India third in 2025. Look again, and the data tells another tale. On supercomputers, India ranks fifteenth out of thirty-one; on compute capacity, twenty-third; on internet speed, thirty-second out of thirty-five.

Without enough homegrown tech infrastructure, we rely on Big Tech, producing little more than AI wrappers. These companies, stacking hollow products atop existing large language models and contributing little of their own, are fated to toll their own death knell.

Take CodeParrot, for instance, which turned design mockups into polished website code by piggybacking on existing models. Then there’s Subtl.ai, a start-up that helped organisations sift through dense archives using, presumably, the same Silicon Valley models. Such ventures featured in headlines like “The AI Startup Dilemma,” often singled out as “mostly an add-on.”

The middleman economy parasitises Big Tech products but also makes the AI landscape in India deludedly robust. On the question of building a model from scratch, Mohit Saxena, co-founder of InMobi and Glance, noted there is little incentive to “reinvent the wheel.” The real intellectual property, he suggests, lies in “data preparation, contextual layering, and post-processing.” A roundabout way of saying it’s an AI wrapper.

This is perhaps why Ankush Sabharwal, founder of CoRover.ai, India’s first “sovereign” conversational AI platform, gets bragging rights. His LLM, BharatGPT, is a “sovereign” large language model, made from scratch. In a recent interview, he even said he wanted nothing less than improving the “ease of living for society.” For all that expansive, aspirational talk, though, his product remains resolutely administrative. CoRover sells AI that automates routine tasks across industry and government, letting users spin up bots that answer customer queries, resolve citizen complaints, and manage internal workflows such as onboarding. Set against general-purpose systems like ChatGPT, they fall short. Tightly task-bound, they cannot take on anything they weren’t made for.

In this way, the language of sovereignty may flatter national ambition, even as the co-dependency on Big Tech remains firmly intact. The middlemen are under no obligation to build large-scale infrastructure. They hover instead in the profitable interstices.

The middleman economy has all the same survived, and it has done so by leaning on a servile class. In their 1995 essay “The Californian Ideology,” Richard Barbrook and Andy Cameron observed that the quest for AI is often a quest for a “strong and loyal slave” to oust an unreliable “underclass.” But it is the informal workforce of this underclass that produces the raw material AI depends on.

In Code Dependent, Madhumita Murgia notes that the push of AI into healthcare has turned India’s ASHA workers into de facto data collectors. Their work now involves “increasingly logging digitized medical information into iPads provided by the government, about everything from household vaccinations to women’s health, children’s nutrition, and sexually transmitted infections.” This information, gathered from poor, rural Indians by these workers – who are often “trusted by local families” – goes on to build expensive technologies for wealthy urban users or Western corporations.

According to Murgia, many of the workers never saw the products built with the data they supplied, and the companies don’t bother following up with them. With the involvement of the AI interloper, even public welfare policy becomes a mine for extraction. No matter that, in redirecting these workers’ labour to extraneous work, the middleman compromises the public service they offer.

All of this works in favour of the “virtual class,” as Barbrook and Cameron describe the techno-intelligentsia who form a new “labour aristocracy.” In line with the middleman economy, this class pockets the profits while remaining removed from the labour that sustains them. Their projects may be presented as collective, national infrastructure, but they pursue “individual self-fulfilment” through the marketplace.

As long as abundant, cheap engineering labour exists, few will reproach the virtual class. All is justified under the guise of providing jobs, and any job is better than the alternative of penury.

India’s AI ecosystem is, for now, a story of middlemen told by middlemen, a tale of progress that is always almost about to touch ground. When the language of “data sovereignty” is invoked to signal a new era of self-sufficiency, it often amounts to little more than a branding exercise, wilfully ignoring just how dependent we remain on the infrastructure of other countries.

True sovereignty, Gupta argues, lies in the “human capital” at the root of the value chain. As it stands, we extract that capital but do not nurture it. He notes, “It is a little-known fact that India leads the world in chip design,” with around 20 percent of the world’s chip design engineers based in the country. “India has long been strong in chip design, but the challenge has been converting that strength into semiconductor manufacturing,” said Stephen Ezell, vice president for global innovation policy at the Washington, DC-based Information Technology and Innovation Foundation.

We appear collared in a self-perpetuating cycle of brain drain. The middleman dream offers “the good life” within India, while those with the most specialised skills increasingly look outward for an even better life in the West.

If AI is to mean anything more than a mythologised extension of India’s feudal past and present, it will require an engineering culture willing to roll up its sleeves and build infrastructure. For now, we remain occupied elsewhere, polishing the tools of the extractor.