The engineering-in-tech leaders who convened in Conrad Bengaluru on January 22 weren't interested in AI hype. They wanted answers to a specific problem: how do you maintain speed and reliability when AI agents write more code than humans can review?
The consensus from CTOs and VPs running thousand-plus developer teams was clear. Code generation has become table stakes. The real bottleneck now sits in the review process.
Swiggy CTO Madhusudhan Rao opened The Economic Times X CodeRabbit meetup by acknowledging that AI-generated code ships with more duplication and other issues than human-written code. His team once saw pull requests (PRs) piling up as engineers scrambled to review output from AI agents. The solution wasn't slowing down code generation, but automating reviews.
Since deploying CodeRabbit as a first layer of review across 250 repositories, Swiggy has processed over 100,000 PRs and flagged 10,000 risks that would have become bugs or security vulnerabilities. The cycle time from raising a PR to getting feedback dropped significantly, because the first review now happens in the Integrated Development Environment. And the number of back-and-forths per PR decreased by 30%.
"Our best engineers are no longer focusing on whether null checks are handled best,” Rao said. “They're focusing on whether this is solving the right problem.”
Following Rao’s keynote, Michael Fox, VP of Go-to-Market at CodeRabbit, demoed the architecture that enabled this shift. Every PR triggers an isolated sandbox that clones the repository, builds a code graph analysis, runs static scanners and linters, and synthesises results through frontier models by OpenAI and Anthropic. The system doesn't just scan changed files, but understands dependencies, impact radius, and how modifications ripple through the codebase.
Then there was the first panel of the day, titled ‘AI-Driven Engineering: How India’s Top Tech Teams Are Rewriting Their Operating Models’. Featuring leaders from BigBasket, Freshworks, and Media.net, it focussed on how AI adoption patterns differ based on organisational maturity and problem domains.
Sarat Buddhiraju, Chief Architect at BigBasket, explained how the company uses AI throughout the product discovery pipeline. Content generation for product images, descriptions, and category tagging is 90% automated. "One problem we deal with is how to make our customers aware of products we have in our assortment," he said. “The next frontier involves using large language models [LLMs] to interpret search queries and match them against catalogue inventory within 500 milliseconds”.
Sreedhar Gade, VP of Engineering at Freshworks, described the company’s platform-first approach to AI adoption across 1,400 developers. Instead of letting engineers pick tools freely, Freshworks built a model fabric service that acts as an intelligent load balancer. "If I really know what tool to use, it doesn't matter. Maybe one year before, context mattered, and most of the tools lacked context. But now, most of them understand that," he shared, describing how AI has levelled the playing field between junior and senior engineers.
Freshworks also built an internal AI playground called ‘Cloud Wars’, where non-technical staff create autonomous agents with zero coding knowledge. Marketing and sales teams have deployed over 150 agents that handle tasks like prospecting, market research, and lead qualification. The company transformed talent acquisition by training an agent to stack-rank 10,000 LinkedIn profiles, filter candidates, and explain its reasoning conversationally.
Akash Agrawal, VP of Engineering at Media.net, emphasised the shift from assistive AI to outcome-based AI in advertising technology. Instead of advertisers manually configuring targeting parameters across terabytes of data, they describe desired outcomes. The system automatically identifies contextual segments, builds audience packages, and predicts campaign performance.
These granular insights were not all. In the fireside chat ‘Systems That Think: The Future of Scalable, Self-Learning Platforms’, RealFast Co-Founder and CEO Sidu Ponnappa and Prabu Rambadran, SVP Engineering, Razorpay, spoke about what constitutes engineering excellence in the age of self-learning systems. As the discussion turned to talent and hiring, Rambadran said the first question he asks in interviews is what candidates think about AI.
"If you essentially say ‘I use AI to write better emails, clean up my documents, and that's pretty much what I've done’, it's a big thumbs down for me," he underlined. Engineering leaders, he opined, should be running personal projects, writing production code, and staying current with tools that change quarterly.
Meanwhile, Ponnappa shared the most aggressive adoption story. His AI services startup mandates that legal, sales, marketing, and recruiting all work on top of the Claude Code stack. A recent sales demo showed a rep building a complete Salesforce integration in 18 minutes based on one customer call transcript. The catch is that successful AI adoption at this scale demands extreme cognitive effort. "This is not the work for just anybody. This is extremely demanding cognitively. And I predict 30-50% of people will burn out due to this," he underlined.
The executives also revealed what keeps them awake at night. For Rambadran, the challenge is redefining performance evaluation when traditional metrics no longer matter. "We are questioning every fundamental that we assume. How you've evaluated people over the last 30 or 40 years is the kind of documents that they've written or the kind of code that they've written. Those things don't matter anymore," he said.
As for Ponnappa: "India now needs to figure out mechanisms to funnel massive amounts of capital into applied AI. Because if we don't, we will lose our number one export line item: IT services.”
The event closed with Nithin Kunimmal, Director & Country Manager - Sales, India & APJ at CodeRabbit, noting that 73% of developers already use AI tools in production, but only 10% use them with proper guardrails.
The consensus of the engineering leaders at The Economic Times X CodeRabbit was that code generation accelerates development, but without review guardrails, it accelerates technical debt and production incidents. The companies winning the AI race aren't those generating the most code. They're the ones that figured out how to review it faster than it gets written.
The consensus from CTOs and VPs running thousand-plus developer teams was clear. Code generation has become table stakes. The real bottleneck now sits in the review process.
Swiggy CTO Madhusudhan Rao opened The Economic Times X CodeRabbit meetup by acknowledging that AI-generated code ships with more duplication and other issues than human-written code. His team once saw pull requests (PRs) piling up as engineers scrambled to review output from AI agents. The solution wasn't slowing down code generation, but automating reviews.
Since deploying CodeRabbit as a first layer of review across 250 repositories, Swiggy has processed over 100,000 PRs and flagged 10,000 risks that would have become bugs or security vulnerabilities. The cycle time from raising a PR to getting feedback dropped significantly, because the first review now happens in the Integrated Development Environment. And the number of back-and-forths per PR decreased by 30%.
"Our best engineers are no longer focusing on whether null checks are handled best,” Rao said. “They're focusing on whether this is solving the right problem.”
Following Rao’s keynote, Michael Fox, VP of Go-to-Market at CodeRabbit, demoed the architecture that enabled this shift. Every PR triggers an isolated sandbox that clones the repository, builds a code graph analysis, runs static scanners and linters, and synthesises results through frontier models by OpenAI and Anthropic. The system doesn't just scan changed files, but understands dependencies, impact radius, and how modifications ripple through the codebase.
Then there was the first panel of the day, titled ‘AI-Driven Engineering: How India’s Top Tech Teams Are Rewriting Their Operating Models’. Featuring leaders from BigBasket, Freshworks, and Media.net, it focussed on how AI adoption patterns differ based on organisational maturity and problem domains.
Sarat Buddhiraju, Chief Architect at BigBasket, explained how the company uses AI throughout the product discovery pipeline. Content generation for product images, descriptions, and category tagging is 90% automated. "One problem we deal with is how to make our customers aware of products we have in our assortment," he said. “The next frontier involves using large language models [LLMs] to interpret search queries and match them against catalogue inventory within 500 milliseconds”.
Sreedhar Gade, VP of Engineering at Freshworks, described the company’s platform-first approach to AI adoption across 1,400 developers. Instead of letting engineers pick tools freely, Freshworks built a model fabric service that acts as an intelligent load balancer. "If I really know what tool to use, it doesn't matter. Maybe one year before, context mattered, and most of the tools lacked context. But now, most of them understand that," he shared, describing how AI has levelled the playing field between junior and senior engineers.
Freshworks also built an internal AI playground called ‘Cloud Wars’, where non-technical staff create autonomous agents with zero coding knowledge. Marketing and sales teams have deployed over 150 agents that handle tasks like prospecting, market research, and lead qualification. The company transformed talent acquisition by training an agent to stack-rank 10,000 LinkedIn profiles, filter candidates, and explain its reasoning conversationally.
Akash Agrawal, VP of Engineering at Media.net, emphasised the shift from assistive AI to outcome-based AI in advertising technology. Instead of advertisers manually configuring targeting parameters across terabytes of data, they describe desired outcomes. The system automatically identifies contextual segments, builds audience packages, and predicts campaign performance.
These granular insights were not all. In the fireside chat ‘Systems That Think: The Future of Scalable, Self-Learning Platforms’, RealFast Co-Founder and CEO Sidu Ponnappa and Prabu Rambadran, SVP Engineering, Razorpay, spoke about what constitutes engineering excellence in the age of self-learning systems. As the discussion turned to talent and hiring, Rambadran said the first question he asks in interviews is what candidates think about AI.
"If you essentially say ‘I use AI to write better emails, clean up my documents, and that's pretty much what I've done’, it's a big thumbs down for me," he underlined. Engineering leaders, he opined, should be running personal projects, writing production code, and staying current with tools that change quarterly.
Meanwhile, Ponnappa shared the most aggressive adoption story. His AI services startup mandates that legal, sales, marketing, and recruiting all work on top of the Claude Code stack. A recent sales demo showed a rep building a complete Salesforce integration in 18 minutes based on one customer call transcript. The catch is that successful AI adoption at this scale demands extreme cognitive effort. "This is not the work for just anybody. This is extremely demanding cognitively. And I predict 30-50% of people will burn out due to this," he underlined.
The executives also revealed what keeps them awake at night. For Rambadran, the challenge is redefining performance evaluation when traditional metrics no longer matter. "We are questioning every fundamental that we assume. How you've evaluated people over the last 30 or 40 years is the kind of documents that they've written or the kind of code that they've written. Those things don't matter anymore," he said.
As for Ponnappa: "India now needs to figure out mechanisms to funnel massive amounts of capital into applied AI. Because if we don't, we will lose our number one export line item: IT services.”
The event closed with Nithin Kunimmal, Director & Country Manager - Sales, India & APJ at CodeRabbit, noting that 73% of developers already use AI tools in production, but only 10% use them with proper guardrails.
The consensus of the engineering leaders at The Economic Times X CodeRabbit was that code generation accelerates development, but without review guardrails, it accelerates technical debt and production incidents. The companies winning the AI race aren't those generating the most code. They're the ones that figured out how to review it faster than it gets written.
(This article is generated and published by ET Spotlight team. You can get in touch with them on etspotlight@timesinternet.in)