Vinayak Jhunjhunwala, co-founder of Bengaluru-based startup Superjoin, has recently been ‘vibe coding’ — giving instructions to AI while it does the coding for him. You don’t have to be a developer or even be familiar with coding; you can just give instructions and an AI tool will do it for you.
In the deeptech space for about six years now, Jhunjhunwala builds data connectors and AI agents for Google Sheets and Excel — has been building projects for his personal use and to use within his company through vibe coding. “I used Cursor to build knowledge-based articles from over 500 conversations we had from customer support interactions. We wanted to create a document that would act as an encyclopaedia for our team. If I had not used vibe coding for this, I would have had to roster a business associate and a tech associate to work together on this, which would have taken about a week. With the help of AI, I was able to complete the task within a night,” he told FE.
The term ‘vibe coding’ was coined by OpenAI co-founder Andrej Karpathy in February, when his post on X took Silicon Valley by a storm. “There’s a new kind of coding I call ‘vibe coding’, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists,” he wrote.
Quite a few tools are already doing this — there’s of course ChatGPT, Cursor, Replit, and Bolt, which are the popular ones, but there are also Anthropic’s Claude, GitHub’s Copilot, Lovable, and Windsurf. Even Google launched Firebase Studio in early April, a cloud-based AI tool that builds apps simply using prompts.
Archit Chauhan, co-founder and CTO of Crib App, a Bengaluru-based property management software company, said his organisation has been using AI assistants like Claude, ChatGPT, and Cursor to assist their engineers in coding. “The workload that needed 4-5 engineers is now being handled by 1-2 at the max. This has increased our speed of shipment and efficiency and has given us time to prioritise other tasks. We’re also not necessarily hiring freshers anymore in our engineering department,” Chauhan added.
However, this could spell a death knell for coding jobs, especially at a time when jobs in the information technology (IT) sector are under pressure. In February, job-hunting platform Indeed reported that openings for software engineers dropped by over one-third in the US from five years ago. A 2022 report titled State of the Developer Nation noted that India has over 5 million active coders employed, the highest in the world.
In India, a 2023 ministry of education report revealed that despite IT jobs enjoying a certain stature and respect, engineering programme enrolment dropped from 4.08 million in 2016-17 to 3.66 million in 2020-21 — a sharp 10% — with some industry experts linking this with the rise in AI development.
A report by Gartner analysts last year had said that over 80% developers would need to pick up additional skills to keep their jobs secured.
All said and done, when tools can code and create apps for you by simply taking instructions, it does create a skill gap in recruitments. “Freshers either need to skill up with AI or increase their efficiency. Senior engineers also need to come in with management skills, etc,” Chauhan said.
Not the endgame
However, while startups are increasingly relying on vibe coding, they don’t think it’s the ultimate solution, rather only a problem solver.
“AI can’t understand the 16-17 legacy codebases we’re using that interact with each other, and its results are not 100% optimal or accurate. What it does best is that it helps reduce the workload,” Chauhan said, adding: “We still need senior engineers to make architectural decisions, to think beyond scale, and to take ownership of modules.”
Rajat Arya, founder of 169Pi, a deeptech AI startup, agreed. “The rise of AI coding tools shouldn’t replace the need to learn programming. It should amplify it. Developers will increasingly be required at later stages where deep expertise, architecture and security come into play.” He believes that while tools like Lovable, Replit and Bolt are helpful in quickly prototyping internal AI assistants and getting ideas off the ground, “they’re not yet robust or built for full, production-level products.”
“If you’re a non-tech person, you would not understand the output that the tool is producing, so no matter how many prompts you give, it won’t be able to match the level of what a developer might create. However, if you are a developer, who has the necessary skills and is using vibe coding to complement their work, then this can help,” said Jhunjhunwala.
But the industry does remain hopeful, with the speed at which AI is advancing, that there’ll be much better versions of vibe coding tools in the coming months.