The term vibe coding was coined by AI researcher Andrej Karpathy in February last year. Photo: iStock
Vibe coding emerged as a phenomenon in 2025, as increasingly capable generative AI tools made it possible to create software by describing what one wanted in natural language. The use has caught on in India, with students to business owners creating bespoke apps for specific needs, cutting down on the expenses that came with employing a coder. But while giving the prompt itself can require some expertise in complex situations, debugging is another area that can leave amateurs stumped.
In September last year, Druthi Polisetty and a sushi-loving friend started Tiny Sushi in Indiranagar, a vibe-filled neighbourhood of Bengaluru, after chasing a “crazy” idea into reality. In the run-up to get the restaurant ready, not only did the two friends try making varieties of sushi — “leading to a lot of rice on the floor” — but Polisetty tried something different. “I vibe-coded the POS (Point of Sale) system because buying it was an expensive option for us,” she recalls.
Vibe coding emerged as a phenomenon in 2025 as increasingly capable generative AI tools made it possible to create software by describing what one wanted in natural language. Without knowing anything about programming or writing a line of code, one can describe an idea to the AI tool and build a custom website or app. With ‘vibe coding’, a term coined by the AI researcher Andrej Karpathy in February last year, by just having an idea and a little patience, functional apps and websites could be built. In a post on the social media platform X, Karpathy wrote, "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...". The term was named its ‘word of the year’ by Collins Dictionary in November. Vibe coding had arrived.
The idea spread in the past year because that is when AI models became better at generating, debugging and modifying codes. Tools such as Bolt and Lovable made software creation accessible through simple prompts. According to data cited by the tech intelligence company Axis Intelligence, the “vibe coding market reached an estimated $4.7 billion in 2026, growing at a 38% CAGR [compound annual growth rate]”. Its growth in the past year is explained by the fact that that’s when AI models became better at generating, debugging and modifying codes. Tools such as Bolt and Lovable made software creation accessible through simple prompts.
That the trend appears to have caught on in India is no surprise. While there is no reliable data available on India’s share of the global vibe coding market, some online sources claim India represents 16.7 per cent of the global vibe coding market share as of the first quarter of 2026, a figure second only to the United States, while others predict Indian developers are projected to become the world's largest developer community by 2027 because of vibe coding. According to the popular 'question-and-answer' website for programmers Stack Overflow’s 2025 global survey, Indian developers show strong adoption of AI tools, with India recording the highest level of trust in AI tools among the survey's top 10 responding countries.
Across the country, vibe coding is attracting entrepreneurs, students and non-technical professionals who want to turn an idea into a working prototype. The uses range from building small apps and websites to automating repetitive tasks and testing business ideas. Meanwhile, for those in the industry, such as software developers, AI coding tools have become part of the development process.
Apps that are vibe-coded as ‘software for one’ bespoke apps mostly solve the coder’s specific issue. Bengaluru-based Nimisha Chanda (24), a marketing professional and co-host of vibe coding hackathons recalls her first product was a personalised GPT to ideate and execute marketing strategies. “Even before ChatGPT I had always ideated on a GPT-like assistant that would serve my purpose,” Chanda says. “I had no idea of coding but was interested in it and wondered how to go about it. I did think of hiring a coder or at least getting help from experts, but those were expensive options. When I heard of vibe coding, I decided to do the app myself. I was motivated enough to study at least 15 AI tools to make one single prototype. Looking back, I believe the FOMO [fear of missing out] of not being in the coding world led me to vibe coding.”
Since last year, she has co-hosted three vibe coding hackathons, where participants use AI tools to build apps, websites, software prototypes or digital solutions.
Before vibe coding, converting an idea into code was an expensive affair, involving proffesional coders. Photo: iStock
The beauty of vibe coding helps create something that is customised to exactly what the user wants and how they want it. From cleaning your phone’s picture gallery, getting curated news bytes to the phone or having an app to recognise the contents in your refrigerator to suggest healthy menus, vibe coding has been one of the ‘hottest’ flavours of the past year.
Polisetty now wants to build an inventory app as well as an app for analytics, which are usually traditionally done by a product manager. “Claude AI [a series of large language models, or LLPs, developed by the American software giant Anthropic] gives me the best codes and metrics,” she says.
Before vibe coding, converting an idea into code was an expensive affair.
Priyanshu Tanwar, a founder (a term used to indicate he is a developer working on a startup) based in Bengaluru, talks of how coding apps or websites have become not only accessible but also affordable now. Before he quit his day job to become a founder, Tanwar was vibe coding a health app called BioPass. Since he was exposed to fitness and wearables and, as a self-admitted fitness freak, he had wanted to make an app that would draw correlations with the fitness parameters, blood profile and so on.
“If not for vibe coding, I would have had to spend at least fifty thousand dollars to create something like that,” he says. But he was doing the app out of “pure self-interest” and gave it up to focus on something completely different, “more B2B focussed”. He stresses the affordability that vibe coding brought to the table. “A whole new world opened up for those who had reasonably good ideas but not the means to code.”
Most AI tools have free and paid tiers. A simple app that might cost Rs 1.5 lakh or more to commission from a developer can potentially be prototyped for free or for the cost of an AI subscription, which typically starts at around 20–25 a month, say users The Federal spoke to.
“To a non-coder, vibe coding seems like sorcery,” says Mumbai-based Akarsh Goel (22), who recently graduated as a lawyer, who “created a website last year to help with studies and to prepare for interviews.” His vibe-coded website was used by his roommates as well, who also helped him tweak it for better results. He spent about four hours of prompting and creating the customised website. Bolt, the AI tool he used, first analysed the task and broke it down into component parts. Then it got to work by generating a basic web interface and for every decision it wanted clarity on, prompted Goel with several options. He recalls, “Every time the tool hit a snag, it would debug its own code or back up to the step where it ran into trouble and try out a different method.”
But that happens only in the case of very simple apps, caution professionals.
Bugs in the code are troublesome to coders, but with vibe coding it somehow becomes more complex, especially when the user has little to no knowledge of programming.
A vibe coder recalls how the app she was working on started to give ‘made-up’ reviews of restaurants. Another user realised how AI was making up legal cases to suit the question asked. As vibe coders soon caught on, AI tools could hallucinate by generating information that sounded plausible but was false. “This is a problem,” Chanda says.
The other problem was to understand where the bug appears and how to solve it, which for a non-technical person would mean mostly staring at lines and lines of code without having a clue.
And so programming knowledge still matters. Knowing a programming language can make it easier to inspect, debug and refine AI-generated code. Chanda says she has learnt Rust, a programming language used to build software, including applications, operating systems and high-performance systems. Even though an AI tool can generate Rust code just as it can generate Python, JavaScript or other programming languages, knowing these languages is helpful. Likening it to a writer asking AI to write a story in a language they do not know, Tanwar believes that if the writer knows the language, they can easily check the story for any inaccuracies or polish it with deeper and more meaningful nuances.
When Polisetty’s POS app had orders disappearing from the to-do list, her background as a software engineer helped her frame questions precisely, allowing the AI tool to identify areas where bugs might have occurred. That is why Tanwar believes that while small apps work well on a simple scale, trying anything complex needs a deeper understanding.
Coders warn that for vibe coding to work precisely in complex scenarios, one has to be clear about the requirements and be very specific with the prompts.
For instance, Polisetty had wanted to view an open order and know what had been served to the table and if there were some additions to that order. “I had to add more features to the app and as the system became more complex, I had to be very careful with giving prompts,” she notes. “As an engineer, I have a robust knowledge of coding, but with vibe coding, Claude did 80 per cent of the heavy lifting and I was able to concentrate on the design,” she says.
Programming knowledged still matters, however, both with giving prompts and with 'debugging' the code when required. Photo: iStock
The art of writing effective prompts has also given rise to the idea of the ‘prompt engineer’, someone skilled at structuring instructions to get better results from A.I. systems. There are now different levels of prompt engineers depending on their ability to structure precise instructions and get the desired outcome efficiently with the least number of tokens (units of text processed by an LLM – a prompt may consist of several tokens). Those who are purely vibe coding (amateurs, or people from non-programming backgrounds, as opposed to coders using AI assisted coding) have realised the importance of precise prompting. Those who are new at vibe coding generally find help on the internet; for instance, Chanda says that whenever she wanted help, she would put out her problem on Twitter and other such platforms. “Someone or the other would always help me out.”
She adds: “If you had asked me two years back, I would have said vibe coding is the relevant stuff, but now I know it is the design that is relevant.” AI finds it harder to understand design, which, she explains, is more ‘taste or aesthetics’.
The distinction becomes clearer with complex software. AI can now generate much of the logic behind an application, but designing a sound architectural foundation remains a difficult task.
Today, LLMs have become capable enough to generate substantial amounts of code, changing how software engineers build and maintain applications. Software engineer Sirisha Annamraju (32) from Hyderabad, who has been in the industry for more than five years, believes this shift is moving engineers to focus more on system design as AI takes on more of the coding. She would still take such codes with a pinch of salt. She warns, “AI-generated code can introduce security vulnerabilities or dependencies that an organisation does not permit and can inadvertently expose sensitive information.”
So what is the future of vibe coding?
Many believe that agentic coding will go a step beyond vibe coding. Instead of simply generating code from a user’s prompts, AI agents can plan a task, write and test code, identify errors and revise the code with limited human intervention. The programmer increasingly acts as a supervisor, like a conductor of an orchestra, setting the goal and reviewing what the agent builds. “If last year was all about vibe coding, now the future may be about agentic coding, where AI not only writes the code, but plans, tests and fixes software with increasing independence,” Goel notes.
According to Tanwar, as AI becomes increasingly capable, it will need to be guided by good thinkers. Whether from a technical or non-technical background, he says, how one thinks could make all the difference to vibe coding.
