Stop Sending JSON to Your LLM Do This Instead

Tool results are one of the biggest hidden costs in AI apps. In this video I show why JSON wastes tokens, how LLM context really works, and how to cut tool result size by up to 80 percent using AI SDK v6 and better output formats. Timestamps 0:00 Why tool results are so expensive 0:41 How stateless LLMs multiply token usage 1:19 The problem with returning JSON from tools 1:52 Returning only what the model actually needs 2:06 Using AI SDK v6 model outputs 2:48 Token friendly formats with Tune 3:34 Returning images directly from tools 4:38 Final thoughts