The rise of artificial intelligence (AI) is reshaping the software development landscape, as evidenced by a notable decline in mobile app-based start-ups and a shift toward AI-powered tools and agent workflows. Recent data from Y Combinator, the prominent Silicon Valley start-up accelerator, highlights this transformation.

Earlier this year, new app releases on Apple’s App Store increased sharply by approximately 80 percent, driven in part by easier app creation tools. Despite this surge, the proportion of start-ups focusing primarily on native mobile apps is shrinking dramatically. Of the 195 companies in Y Combinator’s latest spring cohort, only eight centered their business around a mobile app. This figure contrasts starkly with 2013, when nearly 15 percent of start-ups in the program developed mobile apps. By 2016, that number had dropped to 8 percent and stabilized between 4 percent and 0.5 to 1 percent over the past few years.

The shift reflects broader industry trends, as venture capital and entrepreneurial interest gravitate away from traditional apps and software-as-a-service (SaaS) models toward AI-driven solutions. Start-ups are increasingly focusing on building “agent workflows” and specialization layers over large AI models, such as those developed by OpenAI, Anthropic, or Google. These AI-powered systems automate complex tasks, ranging from financial management to ordering food, by managing AI behavior rather than simply relaying text inputs and outputs.

This new class of software, often called “harnesses,” does more than interface with AI—it controls and coordinates AI responses to execute specific workflows securely and efficiently. For instance, start-ups like Salus and Allowance offer tools to prevent AI agents from making costly errors, such as issuing erroneous refunds or exposing sensitive financial data by providing temporary credit card numbers.

Entrepreneurs like Chaz Englander, who has launched several start-ups through Y Combinator, illustrate this evolution. His current venture enables AI models to perform financial research and report generation by orchestrating multiple AI platforms to contribute their strongest capabilities.

Industry leaders are embracing this model of a centralized AI assistant that can manage various tasks across applications. Meta Platforms recently introduced Muse, a “personal AI agent” accessible through WhatsApp, designed to complete activities like booking travel or sending emails—extending beyond simple information queries into action execution.

This movement toward centralizing capabilities in a single AI agent aims to reduce reliance on myriad apps cluttering users’ devices, offering streamlined efficiency but also concentrating control with AI platform providers. The economic model of these AI-centric start-ups is still evolving and differs from traditional subscription or enterprise pricing structures. Many companies now pursue “usage-based” pricing, charging clients a fee per task completed by their AI-driven systems, purchasing computational resources wholesale from AI labs and reselling the service at a margin.

While the dependency on app stores from Apple and Google diminishes, start-ups now face critical reliance on AI firms like OpenAI and Anthropic. As the sector matures, these changes are poised to reshape how software is developed, delivered, and monetized across the technology industry.