Like many AI-pilled orgs, our spending on coding agents has increased pretty dramatically over the past few months A key part of the LLM gateway we're building (available in private preview now) is cost control and visibility Here's how we're doing it: 1/ Accurate, up to date model pricing. Model pricing is suprisingly non-trivial, with caching, token tier, and other variables coming into play 2/ A lot of integrations. We've had to figure out how to integrate with Cursor, Codex, Claude Code…