started · updated
Nolan Lawson benchmark reveals JavaScript framework efficiency for AI agents
A new benchmark conducted by developer Nolan Lawson reveals that traditional JavaScript framework preferences may not align with the needs of AI agents. Using Claude Opus 5.5 to build a Pokémon application across 12 different frameworks, the study measured development costs, token consumption, time, and performance.
The results indicate that server-centric frameworks are significantly more efficient for AI-driven development than client-heavy Single-Page Application (SPA) frameworks. Hono+htmx emerged as the most cost-effective option with a median cost of $2.67, followed closely by Astro and Nuxt. In contrast, Angular proved to be much more expensive and resource-intensive, with a median cost of $4.97 and token consumption reaching 9.76 million, which is approximately 2.3 times higher than Hono+htmx.
Lawson, a former Microsoft Edge performance project manager currently at Socket, noted that AI agents perform best with frameworks that prioritize convention, API stability, and concision. The study suggests that as AI agents take on more coding tasks, the industry may need to shift its technical selection criteria from human developer experience to model efficiency and token economy.
Entities
Angular · Claude · Microsoft · Nolan Lawson · Socket