Tag
Graham Neubig shares a sidekick architecture for reducing LLM costs by delegating simple tasks to a smaller agent, with a 200-line example using the OpenHands SDK. This approach is also used in Cognition's Devin Fusion hybrid-model harness.