At Alma, an AI-driven nutrition coaching app backed by Menlo Ventures, a striking transformation is taking place: almost all of its code is now generated by artificial intelligence. Rami Alhamad, the cofounder and CEO, emphasized this shift by stating, "I'm not exaggerating. Nearly everything we ship now is AI-generated."
A recent Business Insider survey involving over two dozen startup founders and venture capitalists reveals that AI has swiftly become the primary source of startup coding, with Claude Code from Anthropic being favored overwhelmingly. However, this rapid adoption comes with its own set of challenges, particularly concerns over code quality and potential errors.
The field of coding stands out as one of the most prominent applications for generative AI. Investors are significantly backing AI-focused coding startups like Lovable, Replit, and Cursor, the latter of which was recently acquired by SpaceX for a staggering $60 billion. Furthermore, Anthropic is preparing for a public offering, expected later this year.
Dan Lorenc, cofounder and CEO of Chainguard, an open-source cybersecurity firm, likens the shift to generative AI in coding to transitioning from hand tools to power tools in carpentry. "AI showed up and gave everyone a circular saw," he explained. "It's faster, but there's also a higher risk of mishaps. Right now, many are figuring out how to implement proper safeguards." As of now, Lorenc reports that 100% of his code comes from Claude Code, a considerable increase from 60% just a year ago. He noted that the improvements in models and tools have drastically cut down development times, enabling outputs that previously took weeks to be completed in just a matter of days.
Similarly, Volodymyr Giginiak, CTO and cofounder of Wordsmith AI—a platform for legal professionals—asserted that nearly all of their coding is now AI-generated. "Humans write very little code directly," he remarked, stressing that the critical factor now is not who generates the code, but the level of autonomy granted to the AI. Currently, Giginiak estimates that fully autonomous tasks make up about 10% of the workload but expects that figure to surge to 80-90% within a year. "Engineering isn't disappearing; it's evolving. Those who excel will be adept at creating the right contexts for AI to thrive."
However, this rapid generation of code isn't without its drawbacks. Many founders are reporting a landscape filled with bugs and poorly constructed code. Jason Alan Snyder, a futurist and cofounder of SuperTruth and Artists & Robots, warns of a potential future trend wherein "vibe coding" results in insecure, unmanageable products, developed by teams unable to maintain them post-launch. A report by Menlo Ventures described this phenomenon as the "Cleanup Tax," indicating that while coding speed has increased, the time required for code refinement and quality assurance diminishes return on investment and complicates productivity narratives.
Danny Freed, CEO and founder of Blueprint—an AI platform tailored for therapists—indicated that their reliance on AI-generated code has jumped from 40% in August to nearly full automation today. Freed acknowledges the drastic reduction in costs associated with experimentation but highlights that human expertise remains essential. "Taste and judgment matter more than ever," he concluded, emphasizing that just because something can be created doesn't inherently justify its existence.
