AI is writing almost all startup code. That's creating a new problem.

In today's fast-paced tech landscape, artificial intelligence is revolutionizing the coding process, with many startups relying heavily on AI-generated code. Rami Alhamad, cofounder and CEO of Alma, an AI nutrition coaching app backed by Menlo Ventures, claims that nearly all of their code is now produced by AI. "I'm not exaggerating," he stated. A recent survey conducted by Business Insider reveals that AI tools, particularly Anthropic's Claude Code, have become the primary source of code for numerous startups. However, this rapid development raises concerns for founders regarding code quality and potential issues in the output. While generative AI has emerged as a significant business advantage in coding, with venture capitalists investing billions in AI-focused startups like Lovable, Replit, and Cursor, the speed of generation can lead to subpar results. Notably, SpaceX recently announced plans to acquire Cursor for an astounding $60 billion. Dan Lorenc, cofounder and CEO of Chainguard, a cybersecurity company, compares the rise of AI in coding to the transition from hand tools to power tools in woodworking. "AI showed up and gave everyone a circular saw," he explained. While it allows for faster production, it also introduces new risks. Lorenc noted that 100% of his code is now produced through Claude Code, an increase from 60% last year. He emphasized that while users still need to guide AI, the efficiency gained is monumental. What once took weeks or months can now be accomplished in mere hours or days. At Wordsmith AI, a platform catering to legal professionals, CTO Volodymyr Giginiak echoed similar sentiments, stating that nearly all of their code is AI-generated. He pointed out that the focus is shifting from who writes the code to how much autonomy the AI possesses. Currently, about 10% of tasks are fully autonomous, but Giginiak anticipates that figure could rise to 80-90% within a year. He believes that while engineering roles are evolving, the most effective engineers will be those who can create suitable environments for AI to thrive. However, there are clear drawbacks to this reliance on AI-generated code. Some experts warn of an influx of buggy and fragile products developed by individuals who may lack the skills to support them post-launch. Jason Alan Snyder, futurist and cofounder of SuperTruth, predicts that this trend will lead to a "vibe coding" bubble, where many products are launched without adequate maintenance plans. A December report from Menlo Ventures referred to this phenomenon as the "Cleanup Tax," where the speed of code generation is counterbalanced by the time required for quality assurance and bug fixes. As startups like Blueprint, which has increased its AI-generated code from 40% to nearly 100%, illustrate, the cost of experimentation has significantly decreased. CEO Danny Freed acknowledges the importance of human judgment in this new landscape, stating, "Just because something can be built doesn't necessarily mean it should be built."