Doom scroll: Just how hard will artificial intelligence hit the business models of IT service companies?

The debate has moved beyond incremental automation to deeper concern over AI agents that can generate code, test applications and manage infrastructure. If enterprises trust them at scale, the base of the IT pyramid would shrink, leaving junior coders, testers and maintenance staff adrift.

Revenue growth would slow, margins tighten and entry-level hiring thin out. Equity markets have taken fright. Citrini Research’s latest Substack post rattled investors by outlining how Generative and Agentic AI could pose a structural threat to the offshore labour-arbitrage model over the next 2-3 years. Venture capitalist Vinod Khosla has argued that large swathes of traditional IT services could be automated away.

AI leaders hold the mikes. When Anthropic’s chief Dario Amodei warns that advanced AI could sharply reduce demand for entry-level coders, investors listen.

And with good reason. Just this month, Anthropic sent one shudder after another down IT sector spines: first with plug-ins for its Claude agent capable of handling legal, sales and analytics tasks; then with a code-security tool that identified hundreds of vulnerabilities; and most recently with claims that its models can streamline Cobol code running on legacy systems. Anthropic’s first blow cost investors almost $285 billion in lost market cap globally.

Next, cybersecurity stocks such as CrowdStrike and Okta slid sharply. On Monday, IBM suffered its steepest one-day fall in over a decade as fears grew of AI eyeing old Cobol-heavy systems run by banks and governments. In India, firms like TCS and Infosys employ thousands for mainframe and legacy set-up maintenance.

But is all this anxiety overdone?

McKinsey notes that nearly 70% of the software that powers Fortune 500 companies is two decades old. Updating it is not just a stiff challenge, one must contend with a maze of regulations. While AI may take over some tasks, institutional memory, risk management and compliance oversight may still need human control. Enterprises are unlikely to entrust critical systems to AI agents without guardrails.

Questions of accountability, auditability and cybersecurity loom large. Who bears responsibility if an AI agent misconfigures a financial system? How can AI decisions be explained to regulators? What about data leaks and hallucinations?

AI systems and laws could evolve to allay such worries if the makers of both wrap their heads around what they are dealing with. This may explain why Indian IT service businesses, already well acquainted with legacy software, have tied up with disruptors like OpenAI and Anthropic. Nasscom’s stoic stance reflects this calculus. However, if AI crunches project timelines and clients demand lower bills, efficiency gains may not be able to defend IT margins.

On balance, the truth lies between apocalypse and AI-impact denial. Old business models face structural strain and entry-level roles are likely to shrink. But forecasts of rapid AI adoption ignore legacy systems and regulatory drag. Agentic AI will reshape the economics of IT services, but not overnight. That said, we can expect volatility to persist. The onus is on IT service firms to show how well they can adapt.