Your effort level is STILL TOO HIGH

YouTube · Watch/Read · (gathered 2026-10-14)

• Matt Pocock argues that users consistently set their AI model’s “effort level” too high, mistakenly believing more effort always leads to better outcomes.
• He frames “effort” as akin to adding detail in art or a Minecraft sculpture; while beneficial for complex tasks, excessive effort for simple activities (like small code changes or repo exploration) causes overthinking and token waste.
• High effort levels are counterproductive, burning through tokens unnecessarily, distracting the model, and pushing it towards unwanted outputs, thereby diminishing its effective “juice” in the context window.
• Users should significantly lower their model’s effort setting (e.g., Matt uses Opus 5.5 on medium and considers low) to keep the model smarter, conserve tokens, and preserve context for truly useful tasks.