AI has increased work, not replaced jobs at Policybazaar, says CTO Bibhu Krishna

Policybazaar has continued to hire developers even as artificial intelligence (AI) helps write most of its code, chief technology officer Bibhu Krishna told ET AI, arguing that the technology has expanded the company’s engineering workload instead of reducing jobs.

“About 90% of the code that we generate today is written using AI,” Krishna said. He said AI-assisted development has helped Policybazaar increase its production releases from about 1,500 a month around a year and a half ago to between 3,000 and 3,500 currently.

The increased use of coding assistants has not led the PB Fintech-owned insurance marketplace to reduce its developer workforce, according to Krishna. “We have not asked any developer to go. We have not done that. In fact, we have hired,” he said.


Also Read: Agentic payments could test India’s data-localisation rules: Cashfree’s Reeju Datta

“I think the job has increased, the work has increased rather than replacing or changing,” Krishna said. He attributed the growth in work partly to productivity gains that have allowed Policybazaar to take up projects that teams had earlier postponed because they lacked sufficient bandwidth.

Krishna said developers are now clearing pending tasks, automating more internal processes and building agentic AI systems. “A lot of pending tasks, which were not being done because of lack of bandwidth,” are now being completed, he said.

AI has also made it easier for employees outside the technology function to propose new use cases, adding to the engineering pipeline, Krishna said. “AI has helped in democratising the technology to an extent,” he said, adding that non-technical employees increasingly ask whether teams can automate particular processes.

Developers retain responsibility

Krishna rejected the suggestion that developers are becoming code checkers rather than code writers. “They are still writing code,” he said.

AI can help developers write the integrations, but it cannot independently understand the entire technology environment, he added.

“Unless you have the complete knowledge of the ecosystem, it will be difficult for you to write the correct code,” Krishna said. Without that knowledge, teams could spend more time understanding and rewriting AI-generated code, he added.

If a developer uses a tool such as Claude Code to generate code, the company may use another system, such as the agentic coding environment Kiro, to check it, he said. “We have two different agents probably acting in tandem. One is generating, one is checking,” he said.

This process resolves about 60-70% of potential issues before production, Krishna said. A technology lead or team leader manually reviews the remaining code before the company releases it, he added.

“It’s very easy to generate a code using vibe coding. The difficult part is how to upkeep and do housekeeping of that,” Krishna said.

Human oversight beyond coding

Policybazaar applies the same combination of AI-led execution and human supervision to customer-facing processes, Krishna said. The company uses AI for payments, reminders, collections, KYC verification, document collection, motor inspections and insurance-plan comparisons, he added.

Also Read: AI can speed business loans, but underwriters stay in charge: HyperVerge's Kedar Kulkarni

In motor insurance, AI assesses videos submitted when customers seek to renew lapsed policies and decides in about half the cases whether the footage provides enough evidence or requires a human inspection, Krishna said.

Krishna said the company would retain human oversight even as its AI systems become more capable. “AI can just replace the skills, not the responsibilities,” he said.

The same principle applies when AI recommends an insurance policy or gives a customer an incorrect answer, according to Krishna. “The skills can be any, but responsibility is ours,” he said, adding that Policybazaar remains accountable for decisions made through the AI systems it deploys.

Krishna ruled out removing people entirely from Policybazaar’s operations in the near future. “We need to have a human in the loop,” he said. While AI can identify customer intent and automate routine decisions, it has yet to replicate the empathy required when customers need help during difficult situations, he added.