We used to measure a program by the CPU cycles it consumed, the memory it occupied, and the bandwidth it required. In an AI-assisted workflow, there is a new bill before the program runs: the tokens needed to describe and transform it.
Every symbol travels
Source code is sent to models, repeated across context windows, stored in prompts, and generated again as output. A verbose representation creates cost at every stage. A compact representation can make the same intent easier to move and less expensive to process.
Token efficiency is not just an optimization for a model provider. It can mean shorter feedback loops, smaller build artifacts, lower transfer requirements, and more work completed within the same context budget.
Measure the whole loop
LinkedSpec treats token count as one signal among many. The useful question is not whether a syntax looks shorter. It is whether the complete path from intent to verified execution uses fewer resources while remaining inspectable and safe.