An experiment comparing token costs when using AI coding tools (Cursor) to build the same web app across Rails, Django, and Laravel. Django came out cheapest at 31,301 tokens with zero prompt iterations and no extra dependencies, Rails used 61,141 tokens but produced a more complete setup out of the box, and Laravel was the most expensive at 104,989 tokens due to a debugging session. The post also covers hidden costs like N+1 queries, production issues, and multi-agent token scaling, concluding that the cheapest framework is ultimately the one you can debug fastest.
Table of contents
What Did the Experiment Look Like?Side-by-Side ResultsThe Hidden Cost: What Happens After “It Works”The Real Answer377 Impressions