IIT Madras Warns AI Could Shrink the First Steps of India's Career Ladder

Routine work is disappearing from the training ground

Artificial intelligence was already performing 37% of entry-level tasks in India, compared with 33% globally, according to a 2026 survey of 750 human resources leaders cited in a new IIT Madras white paper. The four percentage point gap puts a concrete figure behind its central concern: the routine work being automated is also the work through which young employees learn a profession.

The paper, How AI Is Reconfiguring Work in India: A Task-Based Framework for Understanding Augmentation, Displacement, New Roles and the Changing Structure of Jobs, comes from the Centre for Responsible AI (CeRAI) at IIT Madras's Wadhwani School of Data Science and Artificial Intelligence (WSAI). It was released at the Industry Conclave on AI Governance 2026 in the week before October 26.

Its warning reaches beyond immediate job losses. Junior programmers learn by writing basic code. New accountants develop familiarity with financial records by processing invoices. Customer service agents build judgment through repeated conversations. As machines take over those assignments, employers face a practical problem: how to develop experienced staff without the same starting roles.

The Cognizant and Pearson survey measures the share of tasks reportedly performed by AI. The white paper provides no estimate of how many entry-level jobs India will lose. Instead, it examines how duties, responsibility and career progression could change across call centres, manufacturing, banking, law, software development and creative work.

Why a task is different from a job

A customer service job contains several activities: answering straightforward questions, checking information, resolving disputes and deciding when a problem needs a supervisor. AI may handle the first two while a person remains responsible for the others.

Professor Balaraman Ravindran, head of CeRAI and WSAI and one of the paper's authors, explained why the researchers start with those individual activities:

Most jobs have tasks and only some of those can be undertaken by AI.

The framework separates four levels: an individual task, the job containing it, the wider occupation and the labour market. Moving from one level to another requires examining what employers actually change, including recruitment, responsibilities and staffing.

Dr Gokul S Krishnan, CeRAI's research director, said the outcome depends on the technology's cost and capability, regulation, safety requirements and organisational choices. A business might use AI to help employees complete more work, redesign their jobs around checking automated results, or reduce demand for particular duties.

The paper groups these possibilities into two broad patterns, with seven pathways within them. In one pattern, AI performs much of the routine work and people verify results or manage exceptions. In the other, people retain the main responsibility and use AI as a tool to work more efficiently.

Why the economics favour routine automation

Indian business process outsourcing (BPO) illustrates the commercial pressure. The paper cites indicative costs of 8 to 20 rupees per minute for a human agent, including the wider costs of employing that person, against 4 to 9 rupees per minute for a managed voice AI service ready for operational use.

Those ranges come from industry pricing estimates. Actual spending depends on the service, its scale and the work required. The ranges overlap, while the lower AI prices help explain why employers handling large volumes of predictable calls may find automation attractive.

Password resets, account balance enquiries, order updates and appointment confirmations are examples of structured work that relatively inexpensive systems can handle. A caller with several connected problems, or a complaint requiring judgment, may still need a human agent.

Ravindran identified BPO as particularly exposed to this pressure:

The BPO industry is something that will be severely challenged by adoption of AI and has been identified as such in the paper.

Which parts of junior jobs are changing?

The paper's sector examples show how routine duties can shrink while responsibility for difficult decisions remains with people. These examples illustrate possible changes rather than measuring the share of work automated in each profession.

• Software engineering: AI can generate basic code, while people handle system architecture, complex debugging and technical decisions.
• Customer support: bots can answer routine enquiries, while agents resolve escalated cases and assess service quality.
• Finance and accounting: AI can process invoices, classify ledger entries, reconcile bank records and prepare information for Goods and Services Tax returns. People verify exceptions and provide professional approval.
• Legal services: AI can assist with research and document review, while lawyers retain litigation strategy, advocacy and client advice.
• Manufacturing: automated visual inspection can flag defects, while workers interpret the findings and manage operational problems.

A smaller bundle of routine duties can therefore leave a more demanding human role. The worker may spend less time processing ordinary cases and more time handling situations where the automated system needs help.

The risk of a missing middle

Junior employees traditionally build expertise through repetition, feedback and gradually increasing responsibility. The paper identifies a possible workforce gap if those starting assignments shrink while businesses continue to need experienced professionals.

Software development makes the problem easy to understand. An employer may still need engineers who can judge whether a system is reliable, secure and suitable for its purpose. Those judgments usually develop through experience with smaller programming tasks, errors and corrections.

The same issue appears in law and accounting. Reviewing documents or checking financial entries can seem repetitive, yet the work helps beginners recognise patterns and spot unusual cases. Replacing those tasks changes the route into more advanced responsibility.

Employers could redesign junior jobs around checking AI output, investigating errors and evaluating performance. Whether companies invest in those career routes or recruit fewer beginners remains undecided. The paper identifies the risk without calculating its eventual size.

Supervising AI requires training too

Universities and employers would need to prepare graduates for work in which an AI system may produce the first answer. A new programmer may be asked to test generated code. A finance graduate may need to verify an automated classification and explain why it is right or wrong.

Those responsibilities require subject knowledge as well as familiarity with AI tools. Checking an answer involves understanding the underlying problem, recognising mistakes and knowing when to seek help. The paper's education concerns centre on how students acquire that judgment.

Ravindran said moving workers into supervisory positions would require training, suitable tools and safeguards:

Job shifts will have to be accompanied by adequate training, development of tools that make it easier for humans to perform the supervisory job and adequate failsafe mechanisms. Again, there is not enough research to answer these effectively as of today. Identifying these problems is the first step to addressing them effectively.

A possible BPO career route could move from customer service into escalations, quality evaluation, AI conversation design and operations supervision. For workers, the practical question is whether those steps come with training and a realistic opportunity to progress.

AI adoption will vary across India

The researchers use the term AI-cost geography to describe how prices and access to technology may shape where automation spreads. Affordable tools can reach routine BPO work, bookkeeping and quality inspection in micro, small and medium enterprises (MSMEs).

Applications requiring more advanced models, specialist expertise or substantial infrastructure, including some legal research and diagnostic imaging systems, may remain concentrated in larger organisations and urban settings. Different kinds of AI can therefore spread at different speeds.

The paper also examines differences between formal and informal employment, urban and rural workers, and small and large businesses. A single national figure would leave out these differences in access, exposure and the ability to move into new roles.

Job quality is another concern. The white paper flags surveillance and work stress, alongside gaps in evidence about informal employment. Measuring the transition means examining pay, workload and working conditions as well as counting positions.

New roles depend on accessible routes into them

NITI Aayog's 2025 Roadmap for Job Creation in the AI Economy projects that India could create up to four million AI-related jobs over five years, conditional on coordinated investment in skills and retraining. The IIT Madras paper cites that projection alongside emerging roles such as AI trainers, conversation designers, ethical AI specialists and remote robot operators.

Workers displaced from existing duties need a route into those opportunities. The paper calls for training tied to realistic alternative roles, with clarity about its duration, cost and whether people can maintain earnings while learning.

Ravindran described responsibility for paying for that transition:

Partly the state, partly the industry and partly the learner – the mix will depend on the industry and wider social impact of the job displacement.

The relevant developments span several stages:

• 2025: NITI Aayog's roadmap set out its conditional projection for AI-related job creation over five years.
• 2026: The Cognizant and Pearson survey of 750 human resources leaders recorded AI performing 37% of entry-level tasks in India, against 33% globally.
• The week before October 26, 2026: The IIT Madras white paper was released at the Industry Conclave on AI Governance 2026.

What the researchers want employers and government to do

CeRAI's recommendations combine continued AI adoption with preparation for changes in work. The researchers call for affordable tools focused on specific tasks, especially for small businesses; more Indian research on affected tasks and skills; and structured routes into changing jobs.

Clear liability rules are also part of the prescription. Businesses and workers need to know who is responsible when an AI system makes a mistake, particularly when a person is expected to approve its output.

The paper describes its classification as a structured hypothesis. Its examples are illustrative, much of the evidence comes from vendors or practitioners, and India has limited detailed research on the combinations of skills that jobs require. It does not quantify total employment, wage or job quality effects.

That research agenda connects directly to training decisions. Employers and educators need to identify which duties are changing, what knowledge the remaining work requires and how beginners can acquire it. For India's new graduates, the immediate issue is access to a first role that still offers a way to learn and advance.

Key Points

• IIT Madras warns that automation could narrow the routine work through which junior employees gain experience.
• A cited 2026 survey put AI's share of entry-level tasks at 37% in India and 33% globally.
• BPO, software, accounting, law and manufacturing illustrate different combinations of automated tasks and human responsibility.
• Technology costs, regulation, safety and employer decisions will shape changes to jobs.
• NITI Aayog's projection of up to four million AI-related jobs over five years depends on coordinated skills investment.
• The paper calls for affordable AI, better Indian labour research, clear responsibility and practical worker transition routes.