AI Revolutionizes Startup Coding, But at What Cost?

⚡

Key Takeaways

1Startups are massively adopting AI to generate code, reducing development time.

2Tools like Claude Code from Anthropic are dominating, but the quality of the code raises concerns.

3Experts predict an increase in fragile and hard-to-maintain products.

💡Why it matters — The rise of AI in coding is redefining the role of engineers, but it poses challenges in quality and maintenance.

AI, a Major Player in Startup Coding

Business Insider conducted an in-depth survey of numerous startup founders to assess the impact of artificial intelligence on software development. The results show that AI has become the primary author of code in many young companies. This rapid evolution presents undeniable advantages but also raises quality concerns.

At Alma, a nutrition coaching app backed by Menlo Ventures, AI is responsible for almost all lines of code. Rami Alhamad, co-founder and CEO of Alma, emphasizes that AI is now at the heart of their development process. "I'm not exaggerating," he states, "almost everything we deliver is generated by AI."

The Rise of AI Tools in Software Development

The Business Insider survey revealed that tools like Claude Code from Anthropic are widely used by startups to generate code. However, this speed of execution comes with concerns regarding the quality of the produced code. Startup founders worry about the waste generated and the reliability of the code.

AI-assisted coding has become a major investment area for venture capitalists, who are pouring billions into companies like Lovable, Replit, and Cursor. Recently, SpaceX announced the acquisition of Cursor for an impressive $60 billion. Additionally, Anthropic is considering an IPO, expected this year.

A New Era for Software Development

Dan Lorenc, co-founder and CEO of Chainguard, compares the use of AI in coding to the evolution of tools in carpentry, moving from hand saws to power tools. This metaphor illustrates the power and risks associated with AI. "AI has arrived and given everyone a circular saw," he explains. "It's much faster, but also easier to get hurt."

Lorenc reveals that 100% of his code is now generated by Claude Code, a significant increase from 60% the previous year. He notes that AI models have improved dramatically, allowing for the production of what used to take weeks in just a few hours.

The Growing Autonomy of AI in Coding

At Wordsmith AI, a platform dedicated to legal teams, Volodymyr Giginiak, CTO and co-founder, indicates that the company's code is almost entirely generated by AI. "Humans write very little code directly," he says. The question is no longer who writes the code, but how much autonomy AI possesses.

Giginiak estimates that fully autonomous tasks currently represent about 10% of the work, but he anticipates a rapid increase in this figure. Within a year, he expects that 80 to 90% of tasks will be fully autonomous. "Engineering is not disappearing, but it is fundamentally being restructured," he adds.

The Challenges of AI-Generated Code Quality

The survey also highlighted a major issue related to the use of AI for code generation: the proliferation of waste and bugs. Jason Alan Snyder, a futurist and co-founder of SuperTruth, warns that the current trend could lead to a wave of fragile and hard-to-maintain products.

A report from Menlo Ventures, an early investor in Anthropic, describes this phenomenon as a "clean-up tax." The speed gains in writing code can be offset by the time needed for cleaning and quality assurance, creating a paradox of return on investment.

At Blueprint, a company developing an AI operating system for therapists, almost all code is now generated by AI, up from 40% last August. Danny Freed, CEO and founder, appreciates the reduced testing costs but emphasizes the increased importance of human judgment. "Taste and judgment matter more than ever," he says. "Just because something can be built doesn't mean it should be."