Zeta’s engineering team has stopped writing code by hand, a shift that signals the end of traditional software development inside the company. Co-founder and APAC CEO Ramki Gaddipati put the question to his engineers recently: how many still open an IDE to type code? None did.
The admission came during the Global Fintech Fest 2026, where Gaddipati told The Economic Times Digital that AI now produces code at Zeta while developers keep full control over every single task. Developers delegate work to AI agents, supply context and instructions, review the output, and remain the only ones authorized to commit and push changes.
The bigger transformation happens during review. Previously, a pull request went directly to a colleague. Now, a suite of AI agents checks the code first for security gaps, runs a full regression suite, and flags sections missing test cases. Only after agents clear the request does a human reviewer see it.
Gaddipati acknowledged that humans scanning three or four pull requests daily often just skim. AI agents, he argued, catch what tired eyes miss. He invoked the banking industry’s four-eye principle, where separate people write and check code. That principle endures, he believes, even as roles merge. With multiple agents acting as dozens of eyes, a single human reviewer could eventually suffice.
Meanwhile, Zeta introduced agent fluency standards three months ago for engineers, product managers, and delivery teams. President of digital banking Sivaram Kowta expects employees to hit baseline within three to four months. He reported a 30 to 40 percent productivity gain and a more than 50 percent drop in staff needed to serve clients. Headcount will hold steady or grow slower as Zeta expands.
Kowta also noted that while roughly 20 banks, mostly Indian, moved AI from pilots to production in the past six months, their underlying infrastructure remains unchanged. AI consumes only 3 to 5 percent of a bank’s net IT spend, funded project by project. That piecemeal approach, he warned, prevents later projects from becoming cheaper or more effective.
Data protection adds another complication. Under India’s DPDP Act, companies can use personal data only for its collected purpose. An AI agent accessing a customer’s balance for one transaction must not reuse that knowledge later. Kowta said legacy banks need stronger AI infrastructure frameworks, which Zeta calls an AI infrastructure fabric.
Zeta also works on agentic commerce, including issuing separate cards to AI agents for user-defined purchases. Gaddipati predicted no Indian launch this quarter or next, citing two years minimum for scale. The US, he said, will lead adoption.
