Nobody at Zeta writes code the old way anymore: co-founder Ramki Gaddipati

When Ramki Gaddipati, co-founder and APAC chief executive officer (CEO) of Zeta, recently asked his engineers how many still opened an integrated development environment (IDE) to write code, the answer was zero.

“Nobody anymore writes code in the traditional sense that I open an Integrated Development Environment, I open a class file and start typing in it. The traditional way of code writing at Zeta is past. It's not going to come back,” Gaddipati told The Economic Times Digital during the Global Fintech Fest (GFF) 2026.

At Zeta, AI now generates code, but developers retain ownership of every task, Gaddipati said. He added that Zeta uses AI generation rather than simple automation. A developer assigns work to an agent, gives it context and instructions, reviews the output first and remains the only person who can commit and push the code.


Gaddipati said the bigger change comes in the second round of review. “Earlier, a pull request went straight to a colleague. Now, a set of agents checks it first for security, runs a full regression suite and flags parts of the code that lack test cases. The human reviewer sees the pull request only after the agents clear it,” he said.

A pull request refers to a request to merge new code into the main codebase.

Gaddipati said AI agents can also catch problems that a human reviewer might miss while scanning several pull requests a day.

“Typically, if you get three, four pull requests every day, you more or less scan through it and say that it looks correct... You’re a human, end of the day. You are not a compiler.”

Banks have long followed the four-eye principle in software development, where two different people handle the writing and checking of code. Gaddipati believes the principle still holds, even though the two roles could eventually merge.

“The maker is seeing [the code] through machines. So the eyes are there. [With multiple AI agents] there are probably 40 eyes now. It comes to me [human reviewer] only after it has been checked by these agents. So, it can be done by one at some point,” he said.

AI agents enter the engineering team

Sivaram Kowta, president of digital banking at Zeta India, said the company introduced a set of “agent fluency” standards three months ago for its engineers, product managers and programme delivery teams. Zeta now measures employees against those standards as part of its competency framework.

Kowta expects the workforce to reach the desired baseline in the next three to four months.

“The company is seeing at least a straightforward 30-40% productivity boost and the number of people needed to deliver services to clients has fallen by over 50%... the headcount will stay the same or grow more slowly than before, since the company is in a growth phase,” he added.

Banks run AI on same old tables

Zeta claims to have spoken to about 20 banks, mostly in India, to understand where they stand on AI. According to Kowta, most have moved from pilots to production in the last six months, “but the underlying infrastructure is still exactly the same”.

Kowta estimates that AI accounts for 3-5% of a bank’s net IT spend. Banks fund one project at a time.

“The Nth project should be cheaper and more effective than the first project. That won't happen if you just take a piecemeal approach,” he said.

Kowta used a bank balance to explain the problem. “Every customer's balance sits in the same database table. An agent that fetches a balance on one customer's behalf needs access to that table, which means it can also reach every other customer's balance.”

The Digital Personal Data Protection (DPDP) Act adds another layer. It restricts companies to using personal data for the purpose for which they collected it.

“The agent is coming to access your balance so that it can do a credit check... Now an agent stores data and has memory. It understood your balance today for the loan transaction. Tomorrow, how do you make sure it doesn't use it for the other transaction?” he questioned.

Kowta said the example shows why legacy financial institutions need a stronger AI infrastructure framework.

Kowta calls Zeta’s answer an AI infrastructure fabric. He described it as a system that lets banks govern, audit and control multiple agents across legacy systems. He acknowledged that some of Zeta’s current AI products will not stay differentiated for long.

Kowta also pointed to Zeta’s voice and text customer-support tool. “The voice and text customer support tool handles close to 80% of support calls for one US client, which runs about 600,000 credit cards on Zeta's platform,” he said.

Kowta declined to name the US-based client.

A payment credential meant for agents

Zeta is also working on agentic commerce, where AI agents make purchases on a user’s behalf. The company is building technology that would let a bank issue a separate card to an AI agent for a purpose that a user sets with the bank, such as buying a particular book on Amazon.

Zeta has not launched the product, and the Indian market may not get one soon.

“I don't think in India somebody is launching a card for an agent in this quarter or the next,” Gaddipati said. “The US, of course, will have a lot of early adopters, because that is still the nucleus of change. There will be early adopters and there is appetite for it.”

On how long the card will take to gain scale across markets, Gaddipati said, “it would be at least a couple of years for scale”.

Most consumers may struggle to trust an agent with a card linked to their broader banking relationship.

Gaddipati identified three requirements for an agent to carry out a commercial transaction: identity, credentials and control.

“Nobody should be given a credential that is reusable... If you give the same credentials to an agent, then once you’ve shared it with the machine, you should expect that it is being replicated at any number of places,” he said.

Kowta argued that only the issuing bank can check whether an agent has stayed within its purpose.

“Visa can’t verify, no network can verify… only the issuer can verify,” he said. If a card meant for Amazon shows up in a Flipkart transaction, “the bank should decline it at the point of authorisation”.

“It's extremely easy to get something plausible from an agent. It's extremely hard to get that reliable, something that will consistently work. That is engineering,” Gaddipati added.

Why Zeta stopped selling to other Indian banks

Bhavin Turakhia and Gaddipati co-founded Zeta in 2015. Zeta began working with HDFC Bank in 2020, after repeated outages prompted the Reserve Bank of India (RBI) to bar the lender from issuing new credit cards and launching new digital products.

For about four years after that, Zeta did not take on other Indian banking clients and instead built its business overseas.

Gaddipati said the company made that choice deliberately. “In 2020, when we started engaging HDFC, it was a very huge transformation project for us... It felt like it was a very big engagement, let's be respectful of what it demands. Don’t dilute our attention, don’t try and chase five other things,” he said.

He said the two companies entered into “a certain kind of exclusivity” so Zeta could stay focused on the projects. By late 2024 or 2025, he said, Zeta had “almost completed everything that we started”.

Zeta now expects a tougher market as bank CIOs deal with DPDP compliance, higher expectations around payment security and new demands from agentic commerce.

“With respect to bank CIOs, I don’t envy them at all. It is not only that they need to learn what it is, they immediately also need to know how to control it,” he said.