Same agent. Same question. Two very different bills.
75% fewer tokens. 72% lower cost. Same model, same prompt. The only thing that changed is the agent could finally see the structure of the codebase instead of guessing from text.
@shiftinbits.com@shiftinbits.com