Vibe coding, the practice of using generative AI to write code, has fundamentally transformed software development in 2025, but the reality is far more nuanced than the hype suggests. Coined by OpenAI cofounder Andrej Karpathy in February, the term became Collins Dictionary’s 2025 word of the year and quickly appeared in job listings across the tech industry.
Sriraam Raja, founding engineer at Decode, exemplifies the complex relationship developers now have with AI coding tools. While he can complete projects twice as fast using AI chatbots, he’s become increasingly cautious about over-delegation. “I was giving away a bit of my agency,” Raja explains, noting that waiting for AI output can disrupt workflow and lead to lengthy review processes. More concerning is what he observes industry-wide: “Everyone’s confidence has increased, but so has their laziness, and their willingness to learn things from first principles has dropped.”
Major tech executives made bold predictions about AI’s coding dominance. Mark Zuckerberg expected AI to write half of Meta’s code within a year, while Anthropic CEO Dario Amodei predicted 90% of code would be AI-generated within three to six months. By spring 2025, AI was already producing about a third of code at Google and on some Microsoft projects, though Amodei’s aggressive timeline hasn’t materialized broadly.
The impact on the job market is significant but complicated. Active job postings for software engineers dropped from 159,000 in early 2023 to 92,500 by November 2025, according to CompTIA data. However, demand for AI skills jumped 53% this year. A Stack Overflow survey found only 19.3% of professional developers don’t use AI, yet less than 3% highly trust it for accuracy.
Quality concerns are mounting. Uplevel’s study of 800 developers found those using GitHub’s Copilot had bugs 41% more frequently and weren’t more efficient or less burnt out. AI-generated code tends to be longer and more verbose, increasing error probability. Tariq Shaukat, CEO of Sonar, notes the industry is moving past “magical thinking” to recognize that while AI produces quantity, determining quality and trust remains difficult.
College seniors studying computer science are now the most pessimistic about their careers, with many regretting their major choice due to AI advances. Yet the field remains divided—43% believe AI will positively impact their careers. The consensus emerging is that well-rounded, creative developers who understand projects holistically will thrive, while those who simply “cleared tickets” face greater displacement risk.
Key Quotes
I was giving away a bit of my agency, and so I made a decision to be very conscious.
Sriraam Raja, founding engineer at Decode, describes his realization that over-relying on AI for coding was disrupting his workflow and diminishing his active engagement with problem-solving, leading him to become more intentional about when and how much he delegates to AI tools.
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. I’ve definitely seen a drop in curiosity that I haven’t seen before, and so that worries me.
Raja expresses concern about the broader cognitive and professional impacts of AI coding tools on the developer community, highlighting a troubling trade-off between efficiency gains and fundamental skill development that could have long-term consequences for the industry.
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.
Tariq Shaukat, CEO of Sonar, captures the central challenge facing the industry as it moves past initial AI optimism: while AI can generate working code quickly, verifying its quality and security remains a significant bottleneck that undermines promised productivity gains.
For a lot of engineers, the only thing that looks different is where they spend their time, not exactly how much time it took.
Amy Carrillo Cotten, senior director at Uplevel, summarizes research findings that challenge the efficiency narrative around AI coding tools, revealing that developers are shifting from writing to reviewing code rather than actually saving time—a fundamental change in the nature of their work.
Our Take
The vibe coding phenomenon reveals a pattern we’ll likely see repeated across AI adoption in knowledge work: initial hype followed by sobering reality checks. The 41% increase in bugs is particularly telling—it suggests AI tools may be optimizing for the wrong metrics, prioritizing speed over reliability. What’s most concerning is the erosion of foundational skills and curiosity that Raja identifies. If junior developers can’t learn through hands-on coding, where will the next generation of senior engineers come from? This creates a potential talent pipeline crisis. The market correction is already visible in Karpathy’s own retreat from AI coding tools and declining traffic to vibe coding sites. The future likely involves a more balanced approach where AI augments rather than replaces human developers, with emphasis on those who bring creativity, systems thinking, and deep understanding—qualities AI still cannot replicate. The real question is whether the industry can course-correct before causing lasting damage to skill development and code security.
Why This Matters
This story marks a critical inflection point in the AI transformation of software development, one of the most scrutinized professions in the automation debate. The gap between executive predictions and ground-level reality reveals important lessons about AI adoption across all industries. The decline in curiosity and first-principles thinking that Raja observes could have long-term consequences for innovation and problem-solving capabilities across the tech sector.
The 41% increase in bugs from AI-assisted coding challenges the narrative that AI automatically improves productivity, highlighting the difference between speed and quality. As companies like Cognition (valued at $10 billion for its AI software engineer Devin) attract massive investment, the security risks and quality concerns could create significant vulnerabilities in critical systems.
For the broader workforce, software engineering serves as a bellwether for AI’s impact on knowledge work. The shift from writing to reviewing code, the displacement of junior developers who traditionally learned on the job, and the potential collapse of career ladders could preview similar disruptions in other professional fields. The 53% jump in demand for AI skills alongside overall job posting declines suggests a fundamental restructuring of tech careers rather than simple replacement.
Related Stories
- PwC Hosts ‘Prompting Parties’ to Train Employees on AI Usage
- The Future of Work in an AI World
- Microsoft AI CEO’s Career Advice for Young People in the AI Era
- Business Leaders Share Top 3 AI Workforce Predictions for 2025
- Bluesky CEO Warns Against Over-Reliance on AI for Critical Thinking
Source: https://www.businessinsider.com/year-coding-changed-forever-silicon-valley-2025-12