Riraam Raja, the founding engineer at the software company Decode, has been experimenting with generative AI for coding for two years. He says using a chatbot lets him complete projects about twice as fast—but only if he codes with intention. One day, he found himself sending instructions to the AI and waiting idly for it to respond. “I realized I could have actively written what I was waiting for the bot to do,” he recalls. “I was giving away a bit of my agency, so I decided to be more conscious about when and how much I delegate.”
Raja is careful now. Relying too heavily on AI can disrupt workflow and create lengthy review processes. He’s also concerned about the broader implications of AI on learning and curiosity. “Confidence has increased, but so has laziness. People are less willing to learn things from first principles, and I’ve definitely seen a drop in curiosity,” he says.
What Is Vibe Coding?
The Collins Dictionary named vibe coding its 2025 Word of the Year. Coined by OpenAI cofounder Andrej Karpathy, it refers to using natural language and AI to speed up coding. Soon after, companies began listing it as a desired skill.
Vibe coding is at the heart of a broader shift in how people approach work in the AI era. Software engineering, long seen as a stable, lucrative career, is now under scrutiny as automation changes how code gets written. Some product managers predict AI will allow them to take on technical coding tasks without engineers, fundamentally reshaping roles.
Executives are bullish. Mark Zuckerberg predicted AI would write half of Meta’s code within a year. Google reports AI was handling about a third of its code earlier this spring. Anthropic CEO Dario Amodei suggested 90% of code could be AI-generated within months, while Cognition, which built the AI engineer Devin, is now valued at $10 billion. Even people with no formal coding training are “vibe coding” their own projects.
The Reality Check
Despite the hype, vibe coding isn’t a magic bullet. AI-generated code can harbor subtle errors and security risks. Overreliance on AI may displace junior developers and disrupt the traditional learning ladder, potentially weakening future talent pipelines. But it also offers opportunities: developers can learn new languages, reduce technical debt, and save time when used judiciously.
Tariq Shaukat, CEO of code verification company Sonar, cautions that while AI produces more code, “it’s getting harder to determine the quality and trust needed to integrate it into your codebase.” A 2025 Stack Overflow survey found that only 19.3% of developers avoid AI, but fewer than 3% fully trust its accuracy.
AI-generated code tends to be verbose, increasing the likelihood of hidden bugs. Amy Carrillo Cotten from Uplevel notes that developers using GitHub’s Copilot weren’t more efficient or less burnt out, and their code contained 41% more bugs. Shifting from coding to reviewing isn’t what many signed up for.
Frank Fusco, CEO of Silicon Society, sees the practical trade-off: “What I used to do in days, I now do in hours using words.” But he worries about the decline in critical thinking and coding fundamentals. “We’re hardwired to find the shortest path, but that isn’t the best way to sharpen skills. Coding is a muscle you have to work constantly.”
A Changing Job Market
AI hasn’t eliminated developer jobs—yet—but the industry is adjusting. Active software engineer postings dropped from 102,000 to 92,500 over the past year, and overall tech job postings fell from 621,000 to 433,500. Yet the demand for AI skills surged 53% this year.
Even computer science graduates are feeling uncertain. A 2025 Handshake survey found many are pessimistic about their careers due to AI advances, though 43% believe AI could have a positive effect. Automation may simply be correcting an overheated labor market: those who focus on creativity and ownership remain valuable, while developers who see their roles as “clearing tickets” may be more replaceable.
Where AI Fits in the Future of Development
AI could expand opportunities for software testers and help reduce technical debt. The most effective use of AI appears to be augmentation rather than replacement. Tim Herbert from CompTIA notes, “Specialized models that handle specific tasks can support human developers without replacing them.”
Security risks remain a major concern, and hype around vibe coding has already cooled. Karpathy himself returned to hand-written code for his latest project after AI tools fell short. If 2025 was the year tech went all-in on AI, 2026 may be the year reality tempers the hype.
For developers willing to stay creative and engaged, AI remains a powerful tool—but the era of vibe coding also serves as a reminder: the human element of problem-solving and critical thinking is irreplaceable.