AI Won’t Create India’s Jobs Crisis. But We’re Looking at the Problem from the Wrong End

A recent opinion article by three Ashoka University economists (titled "AI won’t create India’s jobs crisis, but it could deepen it", published on 16 August in Economic Times) argues that artificial intelligence may not create India’s jobs crisis, but it could deepen an existing one. India already suffers from a shortage of productive middle-skill employment. Too many educated young people compete for too few routes into work that offers status, security and mobility. Much recent employment growth has been in low-productivity agriculture and informal services. AI, by encroaching on the cognitive tasks that once provided entry points into white-collar work, risks hollowing out the middle still further.

The diagnosis is largely correct. The prescriptions — productive non-farm jobs, technology-complementary occupations, and role-specific AI skilling — are sensible as far as they go.

They do not go far enough.

The article treats the problem primarily as a labour-market and skilling problem. That framing is incomplete. India’s AI-and-employment challenge is first and foremost an industrial-development problem.

Employment is a low-hanging fruit 

Economists and politicians naturally focus on employment numbers. Jobs are measurable, politically visible, and directly experienced by citizens. There is nothing wrong with this focus — except when the measurable outcome becomes the driver of public policy.

A government can announce jobs created, workers trained, and certificates issued. The numbers can look reassuring, while the productive structure underneath remains thin. Large numbers of low-productivity jobs plus generic AI-skilling programs do not constitute structural transformation.

India does not merely need to protect an existing stock of employment. It still needs to build a much larger and more productive economy. 

The policy question therefore cannot be limited to “How do we protect employment from AI?” 

It must be: “How do we build the productive economy in which increasingly productive employment can exist in an AI-enabled future?”

Industry precedes employment

Productive employment exists because productive economic activity exists. Enterprises employ people because they have functions that need performing. Those functions change with technology, markets and organisation. Occupations form around the functions; skills form around the occupations.

The correct sequence is:

Industry → enterprise → functions → occupations → skills → employment.

Public policy too often reverses it:

Employment problem → jobs program → skilling scheme → hope for industrial absorption.

That reversal is costly.

Labour-intensive sectors matter. Textiles, footwear, food processing, construction and tourism can absorb large numbers of workers. But labour intensity is not industrial development. A low-technology factory can employ many people while generating little technological capability, weak supplier networks and stagnant productivity. A more sophisticated factory may employ fewer people directly while supporting a larger ecosystem of component suppliers, equipment makers, maintenance firms, laboratories, logistics providers, software companies and technical services.

The objective is not maximum labour intensity at any cost. It is maximum productive employment generated by expanding industrial capability.

This distinction becomes sharper with AI. If automation allows one factory to produce more with fewer workers, aggregate employment need not fall — provided the productivity gains are accompanied by new industries, new enterprises, new suppliers and new occupations. India should therefore stop asking only how many existing jobs AI can preserve. It should ask what new productive systems it can build around the technology.

AI skilling cannot lead enterprise transformation

The original article is right that AI education must become role-specific. But role-specific skilling, offered as a public-policy solution, puts the cart before the horse.

“AI for accountants”, “AI for quality-control technicians”, or “AI for warehouse supervisors” sound practical. In reality, the content of such training depends entirely on how the enterprise is actually transforming: which software it uses, whether its systems are integrated, what data is available, which decisions are being automated or assisted, and what remains the human’s responsibility.

A meaningful course can be designed only after the enterprise has begun deciding how its operations will change. The sequence must be:

Enterprise transformation → changing functions → changing roles → skill mapping → AI skilling and reskilling.

AI is also not a free-floating software layer. In most industrial settings it sits on top of equipment, electronics, instrumentation, sensors, connectivity, computing and data systems. Without access to those layers and a coherent plan to integrate them, AI skilling becomes certification theatre rather than capability-building.

Teaching a worker to use an AI tool inside an enterprise that has not itself transformed is like teaching someone to drive a sophisticated vehicle when the road network does not yet exist.

There is no generic AI-enabled middle class

Discussing “middle-skill jobs” in the abstract is also misleading. A middle-level employee in an IT services firm is not equivalent to one in a chemical plant. A technician in an electronics factory does not perform the same economic function as a technician in mining or logistics. Their technologies, workflows, data, risks and career paths differ completely.

There is therefore no single model of the “AI-enabled middle-skill worker.” There are only sector-specific occupational ecosystems.

In logistics, for example, a warehouse may move from manual inventory to digital systems, sensors, computer vision, robotics and AI-assisted scheduling. The occupational ladder changes with it: trainee → operator → digitally enabled operator → technician → senior technician → supervisor → technology-assisted supervisor. Some tasks disappear; new ones appear. 

The public policy goal should not be to freeze the old ladder in place. It should be to ensure that new, more productive ladders emerge as the underlying industry transforms.

Near-term realities cannot be ignored

One important caveat must be stated plainly. India’s economy is services-heavy. Large parts of IT/ITES, BFSI and business-process work are already experiencing task-level compression from generative AI. That pressure is not waiting for factories to modernise. Cognitive entry-level and intermediate roles in these sectors are being affected now, somewhat independently of the industrial strategy outlined above. Any serious response must therefore address both the longer-term industrial capability gap and the immediate displacement risks in services.

Build layered, sustainable and distributed industries

The longer-term answer remains industrial depth. India needs industries that are layered, sustainable and distributed.

Layered industries develop capabilities across the chain: resources → processing → materials → components → equipment → systems → software → AI-enabled applications → services. The more layers developed domestically, the more opportunities for enterprises, suppliers, technicians, engineers and specialised service providers.

Building multiple layers simultaneously is capital- and capability-intensive. Sequencing and prioritisation matter. India cannot do everything at once. Strategic choices about which layers to deepen first will determine whether the vision remains aspirational or becomes operational.

Sustainable industries develop genuine ecosystems—anchor companies, suppliers, MSMEs, finance, infrastructure and technical institutions—rather than remaining dependent on continuous government support.

Distributed industries allow specialised enterprises, laboratories and service providers to emerge across regions rather than concentrating solely in a few metropolitan clusters or giant factories. The large factory itself may not employ millions, but the ecosystem around it can generate employment across manufacturing, maintenance, logistics, testing, software, compliance, engineering and services. That is how a new occupational middle can form.

Industry associations as capability-builders—with eyes open

Individual enterprises, especially MSMEs, cannot build every required capability alone. Industry federations and sectoral associations could play a larger role—if they are willing and able to evolve beyond their traditional representative and lobbying functions.

Many associations have historically been stronger at advocacy than at deep technology coordination. Turning them into effective capability-building institutions is therefore an institutional challenge, not a simple administrative instruction. Where they succeed, they can help members develop technology-adoption roadmaps covering electronics, instrumentation, software, data infrastructure and automation; facilitate access to credit, subsidies and compliance support; organise shared infrastructure (testing facilities, laboratories, technology demonstration centres, common compute); and coordinate partnerships with universities and technical institutions.

They can also help construct deliberate entry-to-supervisory pipelines: internships → apprenticeships → traineeships → entry-level technical roles → senior technical roles → supervisory and technology-assisted middle-management roles. Young people need opportunities to learn technology by working with it, not merely courses about it.

Government can enable—through policy frameworks, funding and national infrastructure. States and cities can supply land, utilities and local services. Enterprises must still provide the actual workplaces and technology adoption. Associations, where capable, can coordinate. Educational institutions supply knowledge and talent. The result is collective industrial capability-building rather than another standalone skilling scheme.

From defending the old middle to building a new one

AI could deepen India’s existing employment problem. That warning deserves serious attention. But the response should not begin and end with employment numbers or generic AI training.

The deeper sequence is:

Build the industry.

Transform the enterprise.

Integrate the relevant technology stack.

Redesign the functions and occupations.

Then train, reskill and upskill the people who will perform them.

India needs productive non-farm employment, but it needs the industries that generate it. It needs technology-complementary occupations, but it needs technologically transforming enterprises that create them. It needs role-specific AI skills, but those skills must emerge from actual enterprise and industry transformation.

The temptation to optimise for reassuring indicators—jobs created, workers trained, certificates issued—while leaving the productive architecture unchanged must be resisted.

The objective is not to save yesterday’s jobs from tomorrow’s technology. It is to build the industries in which tomorrow’s productive jobs can exist. And if AI is going to transform the occupational middle, India’s answer should not be merely to defend the old ladder. It should be to build a better one.