Write Bullet Points

When working with AI agents on extensive or complex tasks, I’ve found that it’s easy for them to lose focus or drift away from the original objective. For long tasks, a simple technique that tends to improve performance is having the AI create and maintain a to-do list with bullet points about how to implement the big change you asked it to make. I ask it to first write a to-do.txt file with bullet points of all the steps of what they should be doing and then keep updating that file as they go.

This approach serves multiple purposes. First, it forces the AI to break down the task into discrete, manageable steps, which helps with planning and organization. Second, it creates a clear roadmap that both the AI and I can reference throughout the process. Third, and perhaps most importantly, it acts as a set of guardrails that keep the LLM focused on the task at hand.

By bringing the to-do file back into context regularly, we create a mechanism that helps it stay on track. The bullet points serve as constant reminders of what needs to be done, preventing the conversation from veering off in unproductive directions. As each step is completed, the AI updates the file, marking items as done and potentially adding new tasks that have emerged during the process.

This technique is particularly valuable for complex tasks that involve multiple stages or components. It helps maintain continuity and ensures that all necessary steps are completed in a logical order. It’s a simple way to add structure to the interaction and make the AI’s work more methodical and reliable.