An Accel-backed startup CEO says your next user isn't human — and it's changing how software gets built

In a groundbreaking shift, the future of software development is being shaped not for human users, but for machines, according to Guillermo Rauch, CEO of Vercel. During a recent episode of the 'Sequoia Capital' podcast, Rauch emphasized that the focus of software engineering is transitioning away from developers themselves to the AI agents that operate on their behalf. "Your customer is no longer the developer," Rauch stated, highlighting the evolving landscape of technology. As Vercel, a web infrastructure company valued at $3.25 billion in the past year, continues to innovate, the way code is written is changing dramatically. Increasingly, it is designed for AI agents to comprehend and utilize, rather than just for human interaction. Rauch noted that this marks a significant evolution in the industry. He posed the question, "Is there something that I could change about that API that actually favors the LLM being the, quote-unquote, entity or user of this API?" This new AI-centric approach suggests that software tools must adapt to the interactions of large language models (LLMs). As the AI era unfolds, Vercel's user base is expanding beyond traditional developers to include designers, marketers, and even AI agents. These users expect seamless functionality and have little tolerance for errors, which puts immense pressure on product developers. "You want something that works 99.99% of the time," Rauch remarked, indicating the high standards that modern users demand. Last year, Vercel secured $250 million in a Series E funding round led by Accel, with participation from other notable investors like Tiger Global and GV. As AI agents are projected to reshape the internet by 2025, analysts from Bernstein have predicted a future where users may interact with information through these agents rather than directly engaging with apps and websites. However, the integration of AI agents comes with its own challenges. Studies have shown that errors made by these agents can compound significantly, potentially derailing tasks if unchecked. Patronus AI, a startup dedicated to optimizing AI technology, highlighted that even a minor error rate can lead to a substantial risk of failure as tasks progress. Nonetheless, they also noted that implementing guardrails, such as filters and rules, could effectively reduce these error rates, indicating a path forward in harnessing the power of AI agents while minimizing risks.