If AI can write the code, what is left for senior engineers? Meesho’s Kiran Katreddi answers

For engineers and software developers, the path to seniority and career growth was generally a gradual learning curve. Conventionally, the timeline for this growth was dependent on two factors: the number of years spent on the job and solving more problems to master the trade. This enabled them to climb through the ranks one rung at a time.

But, with AI and automation, that timeline is compressing.

Engineers fresh out of college can now work on high-value business processes much sooner in their careers, freeing themselves of the repetitive tasks.


“Engineering seniority is mostly about maturity of developing at scale. That maturity comes by solving more and more problems. Earlier, if a fresher joined as a developer in a company, they solved two, three or five problems in a year. With AI now, everything is automated. They might be solving a lot more problems, let’s say 10 or 20, in a shorter period of time depending on their capability and the opportunities,” Kiran Kumar Katreddi, head of platform engineering and agentic AI, Meesho, told The Economic Times Digital.

Also Read: From 100,000 calls a month to 5 million a day: Inside Vobiz’s AI telephony bet

Katreddi added they are seeing the maturity curve getting shifted much earlier. “In the past if, for example, a junior developer might have taken two years to get to a senior role or even beyond, it is less now.”

He believes the rate of solving problems, too, has multiplied, helping engineers and developers solve more cases. Further, this has led to two things: solidifying the understanding of the system and software development as a whole among the younger hires, and pushing the senior professionals to shift their thinking.

What are senior engineers doing?

According to Katreddi, senior engineers at Meesho are more focussed on systems-design thinking. This allows them to design better architectures of systems that can talk to each other optimally without any friction even at the scale at which Meesho operates.

“The overall goal is accomplished by thousands of micro services working together. That is where the real engineering happens. Code writing is just a very small portion. I think people, at least the senior engineers, are able to spend more time on defining the problems properly,” he added.

The scale of the problems Meesho's engineers are dealing with also puts this shift into perspective.

In a letter to his shareholders in May 2026, Meesho founder and CEO Vidit Aatrey noted that its recommendation system processes billions of data points across more than 100 AI ranking models, trained on 400 trillion-plus input signals, and runs more than 6 trillion inferences every day — each in milliseconds. More than 75% of orders on the platform come through AI-driven personalised feeds.

At this scale, writing code is only one part of the problem. The bigger challenge is designing systems that can talk to each other and continue to work reliably as millions of users interact with them. And that, Meesho says, is where senior engineers are increasingly spending their time.

Another thing Katreddi stressed upon is the scale of experimentation at Meesho. He said AI and automation allows engineers to use their technical judgement to focus more on problems that come with scale. “We are seeing a lot of things getting unlocked because the problems which we kept on pushing for sometime might have now become easy to solve. Why? Because these guys were able to think about it and proactively come back to us with solutions.

He said this is also creating a lot more interesting problems to solve for even junior engineers in the same bucket.

Also Read: In the AI era, your job will be measured by outcomes, not hours worked: Automation Anywhere’s Ankur Kothari

AI is opening up new possibilities for engineers, but whether it genuinely lifts their productivity remains unproven and early research suggests the opposite may be true.

A 2025 randomised controlled trial (RTC) by California-based METR, a research nonprofit that scientifically measures whether and when AI systems might threaten catastrophic harm to society, revealed that when experienced developers used early-2025 AI tools on their own code, they took 19% longer to get the job done. Though the tools felt like they were helping, they actually weren't.

However, Meesho claims to have observed twice the productivity among its engineers with AI adoption. “We don’t measure [productivity] from the lines of code or subjective comments but from the impact engineers deliver. One metric is experiments launched this cycle versus last. With the same number of engineers, we're launching roughly twice as many. Features that took two cycles now take one.”

The 70% AI-generated code story

In the shareholder letter, Aatrey also claimed that 70% of the company’s code is now AI-generated and intelligent systems such as Prism (personalisation and intent-led product discovery) and Chorus (Multilingual AI support for consumers and sellers), among others, are embedded across the software development lifecycle (SDLC).

SDLC is a structured framework development teams use to design, build, test, and deploy high-quality software.

But, what does this 70% figure actually entail? What parts of the code are AI-written and what do coders at Meesho do now when a significant chunk of their job is automated?

Clarifying the number, Katreddi said, “When we say 70% of the code is generated by AI, we're really talking about code generation. Software development is a loop — an inner loop and an outer loop. The inner loop is where a developer builds something, tests it, finds an issue, and goes back and fixes it. That's the part we've automated.”

He said Meesho has built a strong context layer to ensure models receive clean, correct, and bounded inputs. A context layer is an infrastructure component between raw data sources and AI models or agents to manage, filter, and assemble the exact business meaning, memory, and operational rules AI needs at runtime to make accurate decisions instead of relying on static prompts or basic search.

“So, the AI understands what the developer is trying to build, and it covers coding, testing, and code review as a complete loop. That inner loop is where the 70% sits, and that's where we've seen a lot of the benefit,” he added.

Using AI to generate the code is one thing but deployment still needs a human in the loop. At a time when studies highlight the risks of AI agents catching ‘mind viruses’ when they work together, their deployment for mass-scale services still requires checks and balances.

Also Read: AI agents can catch ‘mind viruses’ from each other

“Every piece of code, whether written by a human or by AI, goes through the same review process: a senior developer looking at it, asking the right questions, sending it back with comments, plus security and other reviews,” Katreddi noted.

He further explained that an automated bot catches the obvious mistakes, but the corner cases and production-grade quality checks are still validated by people. “The quality bar hasn't changed — that's why we haven't seen any change in reliability, whether the code is AI-written or human-written.”