Before using an AI agent for a task, ask: 'Could I achieve this goal with a simpler, cheaper method like a well-crafted prompt or basic script?'

The allure of autonomy can lead us to reach for agents when simpler tools would suffice—and be far more cost-effective. An agent's strength lies in handling unpredictable, multi-step workflows requiring real-time adaptation to tool outputs. But for repetitive, well-defined tasks (like formatting data or sending standard emails), the token overhead often outweighs any benefit. Developing this discernment saves significant resources. Start by prototyping the task with direct LLM calls; only escalate to an agent if the task genuinely requires iterative planning, tool use, and adaptation to uncertain outcomes.