Scribd
Upload
8 views

To code is human, to hustle AI_ Demand rises for coders who can evaluate bot-generated code

The demand for coders who can evaluate AI-generated code is rising as generative AI significantly speeds up coding processes, allowing experienced developers to complete complex tasks much faster. Startups are increasingly relying on AI tools to generate substantial portions of their code, with some reporting that AI contributes up to 80% of their codebase. However, this shift presents challenges, as companies must ensure the reliability and accuracy of AI-generated code while adapting to new expectations for software developers.

Uploaded by

kumarin74
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
8 views

To code is human, to hustle AI_ Demand rises for coders who can evaluate bot-generated code

The demand for coders who can evaluate AI-generated code is rising as generative AI significantly speeds up coding processes, allowing experienced developers to complete complex tasks much faster. Startups are increasingly relying on AI tools to generate substantial portions of their code, with some reporting that AI contributes up to 80% of their codebase. However, this shift presents challenges, as companies must ensure the reliability and accuracy of AI-generated code while adapting to new expectations for software developers.

Uploaded by

kumarin74
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

To code is human, to hustle AI: Demand rises for coders who can evalua... https://cio.economictimes.indiatimes.com/news/next-gen-technologies/t...

To code is human, to hustle AI: Demand rises for


coders who can evaluate bot-generated code

Vaibhavi Khanwalkar, Ajay Rag, ET CIO

Nadh explained that complex engineering problems, done through trial and error for days before, now fall
away within hours. Tasks that would have taken a couple of hours, take a few minutes. Fixing obscure
technical issues has become significantly faster.

Until generative artificial intelligence (AI) came into the picture, it was a quiet marathon for Zerodha
every time it launched a new feature. Senior developers measured each line, made it precise and often
rewrote it until it felt right.

Kailash Nadh, the company’s chief technology officer, remembered the time when long hours were
put into generating code. The process was slow, but deliberate. Releases took time, which was fine
because as did perfection.

And then AI showed up.

Nadh explained that complex engineering problems, done through trial and error for days before, now
fall away within hours. Tasks that would have taken a couple of hours, take a few minutes. Fixing
obscure technical issues has become significantly faster.

Another startup founder put this as an “opportunity for a massive productivity boost.”

There is a caveat, though: These gains are best felt by experienced senior developers “who know what

1 of 4 25-Apr-25, 12:31 PM
To code is human, to hustle AI: Demand rises for coders who can evalua... https://cio.economictimes.indiatimes.com/news/next-gen-technologies/t...

they are doing.”

Zerodha is not alone. AI agents are writing significant portions, up to 40-80%, of code across Indian
startups with tools such as ChatGPT, Gemini and Claude, founders told ET. From early prototypes to
production-ready platforms, these AI bots are delivering precise, error-free code, helping teams save
time and dial up efficiency.

This reflects a broader global trend.

More for granular tasks

Tech leaders such OpenAI’s Sam Altman, Salesforce chief Marc Benioff and Anthropic chief executive
Dario Amodei have predicted a future where AI generates most of the world’s code. In March, the
chief executive of Y Combinator noted that around 25% of the startup accelerator’s current startups
were already using AI to write 95% or more of their code.

Nadh explained that rather than using AI tools for generating large swaths of codes, Zerodha is
leveraging platforms like Aider, GitHub Copilot, DeepSeek, Claude and Gemini more granularly, for
tasks like auto-completing boilerplate code, sanity-checking, and making code more concise.

Saurabh Karn, founding member of Sarvam AI, acknowledged the trend and said some of the projects
at the AI startup have 80-90% of the code generated through various agents. “AI may be coming into
the picture, and changing the role of a coder, but the responsibility remains the same: build highly-
compliant, controlled code and models,” he said.

This shift is evident across sectors. Sudip Ghose, founder of luggage firm Uppercase, said AI now
contributes 50-55% to its codebase, up from just 5-10% a year ago.

Inmobi founder and chief executive officer Naveen Tewari took to social media platform X to caution
tech professionals to upskill, as the company expects to write about 80% of its code through AI tools
by the end of this year.

Data management company Reltio, which recently opened an R&D centre in Bengaluru, is generating
about 15% of its code through GenAI, CEO Manish Sood told ET. A third of data security unicorn
Druva’s code is written by AI, ET reported on Tuesday.

This is significantly helping productivity. Ghose said after this shift, initial coding tasks now take
almost 70% less time than before. Sood concurred, saying productivity across different areas due to AI
usage was up about 10%.

Dangers lie ahead

Despite the clear advantages, AI coding has presented its challenges. At e-visa startup Atlys, where

2 of 4 25-Apr-25, 12:31 PM
To code is human, to hustle AI: Demand rises for coders who can evalua... https://cio.economictimes.indiatimes.com/news/next-gen-technologies/t...

25-30% of the codebase is either directly generated or heavily assisted by AI agents, an AI agent once
attempted to simplify a script used for cleaning up old log files. This resulted in a serious error: Due to
a missing configuration, the script deleted an entire working directory instead of just the intended log
files, wiping out essential system files and causing a service to crash repeatedly, said founder Mohak
Nahta.

Erring on the side of caution, online self-publishing and audiobook portal Pratilipi also does not use
AI to write code, but only in basic structuring tasks. “We avoid taking any kind of risk which could be
existential in nature. The bar is high; the code which goes into production has to be error-free. So, we
use AI for structuring and reviewing the code, but not for writing,” chief executive Ranjeet Pratap
Singh told ET.

Adapting to AI

AI’s proficiency in generating boilerplate code and automating repetitive tasks elevates the entry-level
expectations for junior developers. “Entry-level and early-career software engineers will have to
accept and acknowledge AI as a significant and able competitor on the floor,” said Prasadh MS, head
of workforce research at staffing firm Xpheno.

“AI's ability to operate at a scale and speed that are humanly impossible to match is a definite
advantage that enterprises will pay for. It’s natural for enterprises to expect young engineers to come
pre-equipped and prepared to work along with AI agents as digital colleagues,” he added.

Neeti Sharma, chief executive of TeamLease Digital, said startups will eventually not look for software
developers in high volumes. “Companies will not hire coders, but they will look for someone who
knows how to read the code and evaluate its efficacy,” she said.

There are about 1.5 million STEM (science, technology, engineering and mathematics) graduates that
enter the ecosystem every year.

“As seen over the last few quarters, the net number of jobs around data security, AI and machine
learning have gone up dramatically. While coding jobs may slow down, new jobs are coming in.
Founders will now judge a candidate not on the code they have written, but assess their prompt
library,” Sharma said.

By Vaibhavi Khanwalkar & Ajay Rag ,

ETtech

Published On Apr 25, 2025 at 09:40 AM IST

Join the community of 2M+ industry professionals

3 of 4 25-Apr-25, 12:31 PM

You might also like