Summarize this content to 2000 words in 6 paragraphs
(BigStock Image)
There’s a trendy new term going around startup circles that is sparking questions about how AI coding tools could change the way early stage tech companies are built.
Andrej Karpathy, co-founder of OpenAI and a Stanford Ph.D., established the phrase “vibe coding” in a post on X last month. “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.
There are various interpretations of “vibe coding,” but the general idea revolves around using AI tools to write code using simple text instructions, with little human intervention and without full knowledge of what’s going on under the hood.
“I’m building a project or webapp, but it’s not really coding — I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works,” Karpathy wrote.
Engineers debate whether this is an effective way to build software products, and what vibe coding is really all about.
But perhaps the more important takeaway is the power of these tools and the impact on startups.
Companies have been using generative AI to help with coding for several years now, but more advanced tools such as Cursor and Windsurf have supercharged the trend.
“If you’re a startup and you’re not taking advantage of this, you better start experimenting, because this thing is here and it’s real,” said Kevin Leneway, a longtime Seattle-area software engineer. “It’s not magic, but it is very valuable.”
Leaner and cheaper
Leneway, a former startup CTO and developer evangelist at Microsoft, has spent the past year building an AI coding assistant and more recently published videos detailing his experience with Cursor.
Part of Leneway’s current job at Pioneer Square Labs is to help quickly build and test new startup ideas. And AI coding tools have been a key part of his workflow. He said he’s writing three times as much code.
“This is one of those rare opportunities where there’s actually a very big advantage for startups because you can move so fast,” he said. “When you’re starting from zero to one with a new product, that’s where the AI coding, vibe coding, really accelerates things.”
Patrick Ellis, CTO at Seattle startup Snapbar, said his company planned to hire an engineer last year. But once it started seeing efficiency gains from AI coding tools, the startup decided to hold off on hiring.
“As an engineer, I’ve never felt more productive,” Ellis said.
Y Combinator recently polled founders about vibe coding and several cited substantial changes in coding productivity.
Leneway wrote a blog post earlier this month about how startups that embrace AI coding tools and use other new automation capabilities — AI-native startups, in other words — can benefit from smaller, more efficient teams.
Ben Lang, an early Notion employee and current angel investor, noted that “you can build faster leveraging AI and don’t need to raise as much money from VC firms.”
Anysphere, the startup behind Cursor, is an example of this growing trend — taking just 12 months to reach $100 million in annual recurring revenue, and hitting a $10 billion valuation, all with just a small team.
“There could be some incredible startups that would only need two or three developers that are really proficient in these tools,” said Bill Harding, CEO at developer analytics startup GitClear.
Ellis said the AI coding tools are allowing colleagues contribute to software-related tasks, even if they don’t have engineering experience.
Aviel Ginzburg, a Seattle investor and founder of startup community group Foundations, said he’s shipping code for the first time in eight years thanks to vibe coding.
Ginzburg is shocked at the level of skepticism from some longtime tech leaders and engineers.
“I could have at least entertained that six months ago,” he said. “But what has happened over the past several months in terms of acceleration on the code generation side … this is durable. It will change software development forever.”
Limits of vibe coding
There are many unanswered questions about how far a startup can go by only relying on vibe coding, and whether it’s a durable way to build software.
“Zero to one will be great for vibe coding where founders can ship features very quickly,” said Diana Hu, Y Combinator group partner, on a recent podcast. “But once they hit product-market-fit, they’re still going to have really hardcore systems engineering … and you need to hire very different kinds of people.”
Seattle-based GitClear analyzed 211 million changed lines of code from 2020 to 2024 and found a significant increase in duplicate code blocks — a potential downside to relying on AI coding.
“We have much more powerful tools and we can write more code than ever — but it’s that much more difficult to review all of that code, and to ensure that it is consistent enough to be maintainable,” said Harding.
A separate report from Harness showed that using AI-generated code forces developers to spend more time debugging. And a report from Google Cloud found that “improving the development process does not automatically improve software delivery.”
Ellis equates AI coding tools to an electric chainsaws.
“They seems so easy to use and not very dangerous, because they’re not loud, right?” he said. “But they’re very dangerous. It’s still a chainsaw.”
That’s why it’s important for startups to study up on these tools — to help assess their value and any risks.
Leneway cautioned that it’s still super early.
“But we do know things are going to change,” he said. “Being prepared for that is super important, and being open and not just sticking your head in the sand.”
Previously: Is it cheating? AI use during job interviews sparks debate over whether to restrict emerging tools
