AI boosts productivity for few but may hinder most developers, warns Karapetsas.
Buterin finds AI more helpful outside his expertise, less so in domains he knows well.
Effective AI use depends on asking the right questions, not blind reliance.
A growing discussion among leading software figures has raised concerns about how artificial intelligence tools are reshaping developer performance across the industry. While some argue AI is empowering only a small subset of developers, others see its benefits in specific contexts.
Lefteris Karapetsas, a developer and manager, recently stated that current AI tools are disproportionately benefiting a small fraction of programmers. According to Karapetsas, only 5% to 10% of developers are becoming more productive through AI integration, while the remaining majority are becoming dependent on it in ways that reduce the quality of their output.
He warned that up to 95% of developers might now present liabilities in collaborative environments, as they often submit code that is either malfunctioning or dangerously inefficient.
From my experience as both a developer and a manager of developers I have come to think that AI as it is today:– is supercharging 5-10% of developers– it’s dumbing down the rest, turning 90–95% into liabilities for anyone hiring/working with themWhat do you guys think?
— Lefteris Karapetsas (@LefterisJP) June 7, 2025
In his observation, less experienced developers tend to rely heavily on AI-generated code without a deep understanding of its implications. This trend, he noted, often leads to outputs that may function superficially but contain critical bugs or performance issues. He emphasized the risks this behavior introduces in production-level environments.
A Different Use Case From Vitalik Buterin
Ethereum co-founder Vitalik Buterin offered a different perspective based on his recent experience. Buterin said AI tools tend to be most helpful in areas where he lacks expertise. He cited personal examples such as experimenting with native Android development and exploring lower-level Linux internals, where AI assisted him by streamlining research and helping test ideas quickly.
In contrast, Buterin found AI to be less valuable in domains where he holds deep subject matter knowledge. He stated that, in such cases, AI does little to accelerate development or improve decision-making. However, he acknowledged that for experts, AI can still serve as a tool for reducing manual workload and suggesting ideas that may merit further investigation.
Both voices pointed out a major element in maximizing AI’s usefulness: asking precise and informed questions. Citing previous comments by Elon Musk about the Grok AI tool, developers involved in the conversation pointed to effective prompting as a critical skill. Without it, even capable tools can yield suboptimal results.