One of the hottest markets in the artificial intelligence industry is selling chatbots that write computer code.
Some call it “vibe coding” because human software developers work through big ideas. Others dislike the term. However, there is no doubt that these tools are changing the work experience of many high-tech workers.
“The essence is that you’re no longer in the core syntax,” says Cat Wu, project manager for Claude Code at Anthropic. “We’re not looking at every line of code. We’re trying to convey this high-level goal of what we want to achieve.”
However, Wu added that “vibe coding” is not the term she uses. “We definitely want to be very clear at the end of the day that responsibility is in the hands of the engineers.”
Anthropic will launch the latest version of its flagship Claude Chatbot on Monday, boasting that Claude Sonnet 4.5 will be the “world’s best” for coding and other complicated tasks.
The large language models behind generation AI chatbots such as Claude, ChatGpt, and Google’s Gemini are possible, from homework help to organise meal plans, but most companies’ “top use cases” are in coding and software engineering.
“That’s often the case that large organizations get rid of it first,” Walsh said. “I think there’s a wide range of perceptions among these AI model providers that coding is actually the place where it’s most traction.”
And Walsh said that human products are popular with software developers, but he is not the only player in a rapidly growing and consolidating market.
San Francisco and the surrounding Bay Area will remain central to the fight to create the best AI coders. This is not just open rivals and fierce rivals of humanity, but also startups such as Anysphere, Cognition, Harness and Microsoft-owned Github.
“This is the most competitive space in the industry right now,” Windsurf CEO Jeff Wang said on a video call from the startup’s office in Mountain View, California.
Windsurf’s coding assistant was launched a year ago, but was soon discovered at the heart of the bidding war between the high-tech giants as it grew in popularity and hit 200,000 users in its first two months. Openai tried to get it. Google then scoops up Windsurf’s founders and research teams, leaving behind the shell of the company acquired by Cognition, another AI coding startup that it acquired in July.
“It was a really volatile time at Windsurf,” Wang told employees in an email in July, announcing the merger with Cognition, the maker of AI coding assistant Devin. Two months later, the merger between the two companies is “going very well,” Wang told The Associated Press from a conference room called New Kelp City, named after the fictional environment of SpongeBob SquarePants.
Some AI coding assistants automatically terminate code written by human programmers, such as the “AutoCorRect” feature, which suggests the next line of email or text. A more advanced tool known as AI agents gives you more autonomy to access computer systems and do your own work.
In testing prior to the release of Monday, Anthropic said the new Claude Sonnet 4.5 could autonomously code for more than 30 hours on a London-based startup project.
Anthropic’s first coding assistant was developed primarily by chance when the company Boris Cherny began building an internal toy project and using it to accelerate his work. The rest of his team then adopted it.
“As time passed, we realized it was spreading to the virus within humanity,” Wu said.
Humanity said in a consumer usage report earlier this month that coding is the biggest use of Claude, with about 39% of users saying they use chatbots for coding.
In contrast, Openai says writing is the most common work task of ChatGpt, with coding and self-expression becoming more “niche” activities on the platform. Still, Openai aims to catch up, introducing a new GPT-5-Codex in September, which can function longer with complex coding tasks.
Some of the most coveted customers for developers of major AI models are coding startups like Anysphere, the maker of popular coding tool cursors.
In conjunction with Anthropic’s Claude Sonnet, it was performing by Cursor composer and famous AI researcher Andrej Karpathy in February when he coined the phrase “atmosphere.”
“There’s a new kind of coding that I call ‘vibe coding.’ Here we succumb to the atmosphere completely, accept exponential functions, and even forget that codes exist,” he writes in X.
“It’s going to get too good,” he said, so he spoke his instructions, “I barely touch the keyboard,” and used it for a disposable weekend project.
“It’s not actually coding. I’m just looking at things, saying things, running things, copying them. Most of them work.”
A few weeks later, humanity shipped the Claude Code.
Some platforms, such as Sweden-based Lovable, are now compatible with vibe coders with an approach that encourages anyone to “create apps and websites by chatting with AI.” However, most tools are designed for professionals with programming expertise.
This phenomenon has raised the fear of software carriers being unemployed, backed by comments from high-tech CEOs who say AI is speeding up software development and making teams more efficient.
Walsh said Gartner’s position is that AI doesn’t replace software engineers, and in fact they need more.
“There’s a lot of software that’s not being created today because we can’t prioritize,” Walsh said. “So it will drive the demand for creating more software. It will drive the demand for highly skilled software engineers who can do that.”
But the same goes for economists I’m starting to worry that the AI would otherwise have become younger workers and entry-level workers. In a report from last month, Stanford University Researcher “A significant decline in employment for early career workers” (ages 22-25) has been discovered in areas most exposed to AI.
Stanford University researchers also found that by 2024 AI tools could solve roughly 72% of coding problems. It may have been even higher since then.
Karpathy did not respond to requests for comment. But the idea that non-technical people within an organization can “atmosphere” of software that is compatible with business is a misunderstanding of what Karpathy meant when he came up with the term, Walsh said.
“It’s simply not happening. There’s no quality. There’s no robustness. There’s no scalability and no security in the code,” Walsh said. “These tools reward highly skilled technical experts who already know what ‘good’ looks like. ”
Wu told her sister, who is still in college, that software engineering is still a great career and worth studying.
“When I talk to her about this, I say that her AI will make you much faster, but understanding the building blocks is still very important because AI doesn’t always make the right decisions,” Wu said. “In many cases, human intuition is really important.”
