AI is writing almost all startup code. That's creating a new problem. (www.businessinsider.com)

In a striking shift within the startup ecosystem, AI tools — particularly Anthropic's Claude Code — are now responsible for the majority of coding, with some companies reporting nearly all their code being AI-generated. This rapid adoption of AI in coding is seen as a double-edged sword; while it accelerates development and allows startups to test new ideas more readily, concerns have emerged regarding the quality of the code produced. Founders worry that the haste of AI-generated outputs can lead to "sloppy" and fragile products, creating a pressing need for effective quality control and the establishment of "guardrails" to ensure reliability. The phenomenon is underscored by significant investments into AI coding startups and the anticipation of increased autonomy in AI's coding abilities, with projections suggesting that up to 90% of engineering tasks could become fully automated in the near future. Despite the efficiency gains, experts caution that the industry may witness an influx of poorly constructed products that are unsustainable post-launch, a concern that has been dubbed the "Cleanup Tax." As startups navigate this nascent landscape, the emphasis is shifting toward the critical role of human engineers who can ensure quality and context in AI operations, affirming that effective oversight and judgment are more valuable than ever in this evolving coding environment.