Comparing AI Costs Across JavaScript Frameworks
Developer Nolan Lawson tested the cost of having an AI coding agent build the same Pokémon app with different JavaScript frameworks. His aim was to bring evidence to claims that particular frameworks are better suited to agents, rather than judging them solely by developer experience.
The agent built the app from a shared specification using a designated framework and available official tools. Lawson checked conformance, development time, token consumption, and Lighthouse performance and accessibility scores. Most results are medians of five runs; Angular failed to produce a measurable app twice, leaving three runs for some metrics.
AI Coding Costs and Benchmark Limits
Using API prices reported by Claude Code, median cost per run was $2.67 for Hono+htmx, $2.76 for Astro, $3.14 for Next.js, and $4.28 for SvelteKit. Cost ranges overlapped between runs, and the figures do not represent what Lawson actually paid.
The results suggest framework choice can affect agent development costs, but they do not establish a universal ranking. The app is content-focused and lightly interactive; agents did not consistently use available tools, and both generation and Lighthouse measurements varied. Whether the same patterns hold for more complex apps or other models remains an open question.