John Collison notes that developers rarely integrate Stripe's API by hand anymore, delegating the task to coding agents. Consequently, long-standing investments in human developer ergonomics—like detailed migration guides—matter far less. Software providers must now adapt to AI ergonomics, focusing on machine-specific challenges such as preventing agents from integrating incorrect API versions.
Instead of waiting years for legacy public and private institutions like DMVs to build modern APIs or protocol servers, computer use allows AI agents to interact with existing web interfaces. John Collison argues that computer use functions as the ultimate backwards compatibility layer, enabling autonomous personal agents to execute real-world tasks without requiring backend modernizations from legacy organizations.
AI-assisted product research serves as a new aggregation layer that bypasses traditional retail gatekeepers. Because agentic discovery naturally uncovers well-reviewed niche brands from the long tail rather than defaulting to dominant incumbents, John Collison predicts a major turnover among top merchants, favoring businesses that adapt natively to AI-driven acquisition.
Modern software products increasingly incur variable compute and inference costs for research and product delivery, making static subscriptions economically fragile. John Collison explains that nearly all AI products must introduce usage-based components, credit systems, or overage charges to align consumer pricing directly with underlying model expenses.
As advancing AI models improve adversary capabilities, John Collison expects security breaches to rise over the next five years. However, startups hold a structural defensive edge: their homogenous, modern tech stacks allow them to deploy the latest AI security tools rapidly, whereas legacy enterprises remain vulnerable due to 50-year-old software and unsupported legacy operating systems.
Traditional web infrastructure relies heavily on bot detection to block automated interactions. As consumers increasingly deploy personal agents to conduct research and purchases, internet traffic will become dominated by bots. John Collison highlights that website operators will soon need new screening paradigms that distinguish helpful customer agents from malicious threats.