Amjad Masad, CEO of Replit, delivered a stark prediction at Y Combinator's AI Startup School: the value of all application software will eventually "go to zero." This bold statement underscores a profound shift he sees in how software is created and consumed, driven by the rapid advancements in artificial intelligence. His commentary, delivered at the "AI Startup School" event on June 17, 2025, traced the historical arc of computing and proposed that AI agents are ushering in an era where programming, once the domain of experts, becomes universally accessible, fundamentally reshaping business and employment.
Masad drew a compelling parallel to the evolution of computing itself. Mainframes, he reminded the audience, were complex machines requiring specialized experts to operate. Personal computers, initially dismissed as mere toys, eventually became indispensable tools for everyone, powering the global economy and even modern data centers. Software engineering, historically demanding years of specialized training and arcane knowledge, is undergoing a similar democratization. As Masad put it, "Software engineering for experts. Replit for everyone."
Replit's nine-year vision has always been to make programming accessible to all. With the advent of AI, this mission has found its ultimate expression: eliminating the need for manual coding entirely. "Code is the sort of bottleneck to actually getting a lot more people making software," Masad explained, highlighting the strategic pivot towards AI agents as the definitive solution. This represents a fundamental re-evaluation of where value is created in the software development lifecycle.
The rapid increase in AI agent capabilities is evident. Masad showcased a steep upward curve on the SWE-bench, a software engineering benchmark. While agents "barely worked" in 2022, they are now achieving 70-80% success rates, demonstrating a clear trajectory towards automated software engineering. Replit has gone "all in on agents," recognizing this inflection point as the moment for significant investment and innovation.
However, the real challenge isn't simply generating code, but building the robust infrastructure for these agents to operate effectively. "Code is easy, infra is hard," Masad stated, emphasizing the critical need for a stable and comprehensive "habitat" where agents can thrive. This includes sandboxed virtual machines to prevent agents from compromising local systems, scalability to support millions of users, and universal language support to accommodate diverse programming environments. Agents need the ability to interact with the command line, manage files, install packages, and handle complex integrations, mirroring the comprehensive toolkit of a human developer.
