Sriraam Raja, the founding engineer at the software company Decode, has been using generative AI to write code for two years. He says he can get projects done about twice as fast when he uses a chatbot to code with intention. Then one day, he fired off directions, and as he sat there while the bot’s wheels turned, he realized he could have actively written what he was aimlessly waiting for the bot to do. “I was giving away a bit of my agency, and so I made a decision to be very conscious,” he tells me.
Raja has become “very specific about when I delegate, and also how much I delegate,” he says. Waiting for the AI to spit out code can disrupt the flow of his work, and trusting too much work to it has led him to sometimes get bogged down in a lengthy review process. He’s also anxious about the long-term effects AI can have on how we all think and problem solve. “There’s a side effect where everyone’s confidence has increased, but so has their laziness, and their willingness to learn things from first principles has dropped,” he says. “I’ve definitely seen a drop in curiosity that I haven’t seen before, and so that worries me.”
The Collins dictionary made vibe coding its 2025 word of the year. Coined by OpenAI cofounder Andrej Karpathy in February, the term refers to using language and generative AI to speed up the coding process. Soon after, companies were adding it as a desired skill in job listings.
Vibe coding was the catalyst for the sort of vibe work era we’ve entered. It’s a shift in how people think about their roles and relationships to work amid an AI boom, and software engineering, long considered a stable and lucrative career path, has perhaps been the career most scrutinized and pushed down a path toward automation. Product managers have suggested that AI will supercharge them, allowing them to take on some technical coding tasks and work without engineers.
Execs have been all-in: Mark Zuckerberg said he expected AI to write half of Meta’s code within a year; this spring, AI was already doing about a third of code at Google and on some Microsoft projects. Anthropic CEO Dario Amodei predicted in March that 90% of code would be generated by AI in three to six months. The bullish estimate hasn’t materialized for most, but Amodei said in October the company’s AI tool Claude was writing most of the code at Anthropic. Cognition, which built an AI-powered software engineer it named Devin, is now valued at $10 billion. Some without computer science backgrounds or any training in coding are vibe coding their own projects.
Vibe coding isn’t yet the miracle that AI evangelists have professed. AI-generated code can have sneaky errors that pose security risks. As it takes on the work of junior developers, companies eager for gain could displace humans. Time banked with shortcuts now could disrupt training ground for learning basic coding skills, creating a tech worker career ladder collapse could ricochet through the industry. There is potential for developers to save time, to use AI to learn new languages and skills (something Raja tells me he’s done), and to pare down their technical debt, or code that needs maintenance. But the impact of AI on the industry is more complicated than it is a silver bullet to efficiency.
Last year, “we were dealing with a lot of optimism and a lot of magical thinking” around the capabilities of AI, says Tariq Shaukat, CEO of Sonar, a company that provides developers with tools to verify code. “The vibe engineering tools are producing a lot of quantity. It’s getting more functionally correct, but it’s actually becoming more difficult to determine the quality and get the level of trust that you need to integrate that into your code base.” The ranks of AI holdouts among developers are shrinking. A 2025 survey of professional developers from Stack Overflow found that only 19.3% don’t use AI, and a commensurate 19.7% have an unfavorable opinion of AI. Yet less than 3% of respondents said they highly trust AI for accuracy.
Anyone who has asked a chatbot a question knows that even a short inquiry often results in a verbose response. The same is true of code — when AI generates it, it’s typically longer, making the possibility of errors hiding in the code more likely. Amy Carrillo Cotten, senior director of customer transformation at software development company Uplevel, told me in September: “For a lot of engineers, the only thing that looks different is where they spend their time, not exactly how much time it took.” Uplevel studied 800 software developers last year and compared the productivity levels of those who used GitHub’s Copilot to those who did not. The developers who used Copilot weren’t more efficient or less burnt out, and their code had bugs in it 41% more frequently. (GitHub’s own research found that those who used Copilot wrote about 18 lines of clean code, compared to 16 lines for those who didn’t.) For many, that shift from writing to reviewing code is “not the job they signed up for,” Shaukat says, which brings a big adjustment for many developers.
“The job looks completely different,” says Frank Fusco, CEO of a software company called Silicon Society. His company works with clients on their software, but now they often get amateur, vibe coded versions of those ideas as the starting point. “What I would normally do in code that would take me days, I now do in words and it takes me hours.” But Fusco tells me he worries about a decline in critical thinking and basic coding skills. We’re “hardwired,” he says, to find “the shortest path to the solution.” But that approach isn’t the best for sharpening coding skills. “It really is a muscle that you have to work all the time.”
It’s tricky to say AI is already killing developer jobs. Years of layoffs and “right-sizing” in the tech industry, paired with the economic precarity that has also defined 2025, could be shifting industry roles alongside AI. As of November, there were about 92,500 active job postings seeking software engineers, down from nearly 102,000 last November and 159,000 at the start of 2023, according to data from CompTIA, a nonprofit trade association for the US IT industry. The number of active tech job posts overall has fallen, from 621,000 in early 2023 to 433,500 last month. But the proportion of open jobs looking for AI skills has jumped by 53% this year.
After two decades of being told to pursue computer science as a stable career and a proliferation of coding bootcamps, working as a developer may not be as cushy. College seniors studying computer science are more likely than any other discipline to say they’re “very pessimistic” about their careers, according to a 2025 survey from early career website Handshake. They’re the group most likely to say the advances of generative AI have made them regret their major choice. But young people are divided — 43% of computer science majors said they think AI will have a positive effect on their careers.
Automation is in some ways marking “a correction” on the developer labor market, says April Schuppel, developer relations manager at software company Apryse. Before AI, “we needed as many people who were really pushing out the code to take the ideas of the visionaries and bring them to life.” Now, “the people who have always been able to make the most impact, they’re still the ones that are the safest.” Developers who looked at their jobs as clearing tickets might be more replaceable than those who were creative and cared about the project from start to finish. We’re far from realizing the end game of vibe coding, but for creative, forward-thinking developers, there’s optimism for now. “The more well-rounded people are the ones that are going to have success,” Schuppel says.
AI could bring more opportunity for software testers, and also help companies pare down their technical debt. The developer job market might look messy right now, but there’s still a heavy focus on the human aspect of the career than in the picture painted by some Big Tech execs. “If there are opportunities for more fine-tuned models, more specialized models that only do certain types of code updates, and there is a way to use that more to augment human developers as opposed to replace, that seems like that’s where this is going,” says Tim Herbert, chief research officer at CompTIA.
Codebases are valuable, and the security risks posed by goofs in AI code are serious threats. Traffic to vibe coding sites slumped in September after a summer of hype. Even Karpathy said his latest project is “basically entirely hand-written (with tab autocomplete)” in a post on X. “I tried to use claude/codex agents a few times but they just didn’t work well enough at all and net unhelpful.” If 2025 was the year tech companies went all in on AI, 2026 might be the year when some of the craze around vibe coding subsides and reality sets in.
Amanda Hoover is a senior correspondent at Business Insider covering the tech industry. She writes about the biggest tech companies and trends.
Business Insider’s Discourse stories provide perspectives on the day’s most pressing issues, informed by analysis, reporting, and expertise.
