IBM Software India General Manager Sriram Raghavan told The Hindu Business Line that AI should be regulated according to its use cases, rather than through one overarching AI-specific law. He said rules will differ across countries and industries, noting that healthcare data should not be treated the same way as retail data. For banking, he pointed to the Reserve Bank of India’s updated model risk management laws as an example of existing regulation being adapted to AI. He said regulators should focus on how a technology is used and extend relevant laws to address its effects, rather than regulate mathematics or matrix multiplication. He added that consumer technologies offered to the public at large may warrant horizontal safety considerations.

Raghavan said AI is changing software development and described AI agents as a new class of applications emerging alongside shifts brought by the internet, mobile technology and AI. Building agents with AI models introduces imprecision, he said, so testing, security, safeguards and measurement need to evolve. He said AI has improved the rate and pace of software work, with economic effects extending beyond productivity to faster delivery of new features and capabilities to customers.

To prepare software engineers, Raghavan called for changes to curricula and for faculty and students to receive AI-native tools before entering industry. IBM is taking its agent Bob to academic institutions, he said. On cybersecurity, he said vulnerabilities and fixes are being discovered at a dramatically higher rate, making prioritisation and automation key challenges for clients. Manually applying every patch would not be feasible, he said; mission-critical systems need extreme measures, while broad use of automation is essential. “If bad guys use AI, you have to use AI,” he said.

On AI costs, Raghavan said discussion has moved beyond token pricing toward optimisation. He said frontier models cost “so many dollars per million tokens” and described choosing a combination of frontier and open-weight models for agents or applications to avoid dependence on one model. He said those choices lead to questions of sovereignty.

IBM Software India General Manager Sriram Raghavan told The Hindu Business Line that AI should be regulated according to its use cases, rather than through one overarching AI-specific law. He said rules will differ across countries and industries, noting that healthcare data should not be treated the same way as retail data. For banking, he pointed to the Reserve Bank of India’s updated model risk management laws as an example of existing regulation being adapted to AI. He said regulators should focus on how a technology is used and extend relevant laws to address its effects, rather than regulate mathematics or matrix multiplication. He added that consumer technologies offered to the public at large may warrant horizontal safety considerations.

Raghavan said AI is changing software development and described AI agents as a new class of applications emerging alongside shifts brought by the internet, mobile technology and AI. Building agents with AI models introduces imprecision, he said, so testing, security, safeguards and measurement need to evolve. He said AI has improved the rate and pace of software work, with economic effects extending beyond productivity to faster delivery of new features and capabilities to customers.

To prepare software engineers, Raghavan called for changes to curricula and for faculty and students to receive AI-native tools before entering industry. IBM is taking its agent Bob to academic institutions, he said. On cybersecurity, he said vulnerabilities and fixes are being discovered at a dramatically higher rate, making prioritisation and automation key challenges for clients. Manually applying every patch would not be feasible, he said; mission-critical systems need extreme measures, while broad use of automation is essential. “If bad guys use AI, you have to use AI,” he said.

On AI costs, Raghavan said discussion has moved beyond token pricing toward optimisation. He said frontier models cost “so many dollars per million tokens” and described choosing a combination of frontier and open-weight models for agents or applications to avoid dependence on one model. He said those choices lead to questions of sovereignty.