Nolan Lawson evaluated twelve JavaScript frameworks by using AI agents to build the same application, finding significant differences in the median JavaScript payload size between frameworks. While AI execution costs remained largely overlapping across stacks, the research highlights how framework choice dictates the weight of client-side code delivered in agent-generated projects.
Key points
AI agent execution costs do not vary significantly between common JavaScript frameworks.
JavaScript payload sizes differ drastically, ranging from 0KB for Astro to over 140KB for heavier frameworks like Next.js.
Developers can use benchmarking tools to evaluate which stack provides the most efficient output when relying on AI-driven code generation.
Framework choice remains a primary factor in application performance regardless of whether code is written by humans or agents.