How I Gave My AI Clear Roles Instead of Tasks
I used to give my AI tasks. “Summarize this.” “Write that.” “Fix this paragraph.” It worked—but it always felt brittle. The change happened when I noticed something: every time results were off, it wasn’t because the task was wrong. It was because the role was unclear. So I stopped assigning tasks and started assigning jobs . Instead of telling Copilot what to do, I told it who to be. Reviewer. Architect. Translator. Critic. Suddenly, the output snapped into place. The insight was simple: Tasks tell AI what to do. Roles tell it how to think . And how it thinks determines the quality of the work. Define the role first. “You are a critical reviewer” changes everything. Match the role to the moment. Planning needs an architect. Polishing needs an editor. Keep roles consistent. Same role = predictable results. Switch roles deliberately. Don’t mix brainstorming and judging. Name the role out loud. It forces clarity before the work starts. Once I made this shif...