AI writes 86% of Paisabazaar’s new code, but CTO says engineers aren’t going anywhere

Artificial intelligence (AI) now writes 86% of the new code at Paisabazaar, but the financial marketplace has no plans to trim its engineering workforce. Instead, it is expanding the technology team as AI allows engineers to take on a larger backlog of products, chief technology officer (CTO) Mukesh Sharma told the Economic Times Digital.

“I am proud to say almost 86% of our new code is now powered by AI. Everyone in my engineering team is on the enterprise version of Claude or Cursor.”

The productivity gains are already showing up in how frequently Paisabazaar pushes software into production, Sharma said. “If in a month I used to do 30 builds on production, it is now more than 150 on production.”


While Paisabazaar is using AI for code reviews, quality assurance and generating test cases, the increase in AI-generated code has not translated into fewer engineering roles. Sharma said the company has more products to build than its existing teams could previously handle.

“If I have 10-12 products in the pipeline that I want to build and ship. But I cannot hire more people endlessly because I have budget constraints,” Sharma explained. “With AI, I am able to ship those faster.”

Also read: Fintechs leverage AI to boost creditworthiness and reduce rejections

“So Paisabazaar is, in fact, beefing up its technology team. We are not cutting people,” he added.

Sharma expects the impact of AI on engineering headcount to vary depending on the type of organisation. Companies largely maintaining mature systems may be able to operate with smaller technology teams, while businesses continuously developing new products will still require engineers, he said.

“Organisations where new development is not happening and they are maintaining an existing system, can cut because you don’t need so many people to maintain an existing system,” he said. “But in an organisation where every day there is new growth and new projects are spinning up, you will need developers and AI.”

The engineer’s job is changing

While AI may not be reducing the size of Paisabazaar’s technology team, it is changing what engineers are expected to do.

Engineers need to understand an existing system, break a technical requirement into smaller tasks and decide which AI model is appropriate for each job, Sharma said. “The job of an engineer is to efficiently use AI to improve productivity. It is as simple as that.”

For a complicated architectural change, for instance, a senior engineer could use a more capable reasoning model to develop a plan. Smaller and cheaper models could then be used to execute individual parts of it, he said.

“You don’t give everything to the world’s best model even if you’re writing a single line of code. That’s overkill. That’s overengineering,” he said.

Sharma, however, flagged the downside for junior engineers, who become too dependent on coding assistants.

“Think of it from the perspective of a fresher who joins my team. From day zero, his code can look like the code of a principal engineer who has spent 12-15 years in the industry. That’s the beauty of it,” Sharma said.

Also read: In financial services, trust comes before speed; AI doesn't replace it

“Where it hurts is that, as a junior engineer, if you constantly take help from AI, you are losing your natural ability to understand the code and understand the basics of the technology.”

AI touches core code, but cannot ship it

Paisabazaar is allowing AI-generated code into some of its core systems, although the company restricts which parts of its technology stack can be exposed to external AI tools.

“AI does touch our core systems. However, they are ring-fenced and we decide what it can touch,” Sharma said.

One area that remains off limits is Paisabazaar’s automated speech recognition, or ASR, technology. “Our ASR models are among the most important things for us. We are not sharing those with AI. There is no way AI is touching those systems,” he said.

AI-generated code also cannot move directly into production. A developer first reviews and commits the code, after which it goes through the company’s existing code review, automated testing, quality assurance, security checks and user acceptance testing processes, Sharma said.

Also read: Fintechs want to bank on agentic commerce

Coding tool to revenue engine

Beyond engineering, machine-learning models now touch at least 50% of Paisabazaar’s revenue, Sharma said. According to him, a large part of this comes from the company’s “Chances of Approval” initiative, which uses Paisabazaar’s credit data and its understanding of lenders’ policies to rank the loans for which a customer is most likely to qualify.

“That initiative now contributes close to 50% of Paisabazaar’s revenue,” Sharma said.

The company is also deploying AI across its call centre, which has close to 3,000 agents, and records around 100-120 GB of calls each day. It then uses in-house speech models to transcribe and analyse them for fraud, do-not-disturb requests, sentiment, compliance and whether agents are pitching appropriate products, he added.

“Earlier, when we did quality assessment of these calls, the best-case assessment coverage was around 25%,” Sharma said. “But now we are doing 100% call audits, in near real time.”