video1.0<iframe src="https://www.loom.com/embed/83ecfb95ebe944b0a267235a94adcebd" frameborder="0" width="1108" height="831" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>8311108Loomhttps://www.loom.com8311108https://cdn.loom.com/sessions/thumbnails/83ecfb95ebe944b0a267235a94adcebd-07e9a4399700c2a9.gif177.348This Loom explains a cost-aware coding agent approach to stop agents from routing every task to the same flagship model. It describes a pipeline with Planner and Schedule, adding a Debate Layer where multiple agents discuss and narrow each user prompt to two different models from a list of five. The system then runs the task on those models, returning generated code along with an exact cost breakdown and critic-based scoring on code quality and performance. The creator notes added latency because two models are used, and mentions Sonnet, Opus, GPT-Mini, and GPT-Nano, with plans to add more models later.