I am gradually developing my own way of working with AI: not as a search box expected to return one final answer, but as a collaborator placed at different stages of a task and continuously challenged and calibrated.
My current division of work
- Exploration: expand the question and expose missing perspectives without deciding the goal for me.
- Research: organize material and surface contradictions while important facts are verified against primary sources.
- Planning: propose alternatives, reveal assumptions, and compare the cost of different paths.
- Execution: handle structured and repetitive work while I review boundaries and final quality.
- Reflection: analyze failures and turn useful lessons into reusable prompts, rules, and checks.
What I will add next
I will add examples from real tasks: what to delegate, how much context to provide, when to stop prompting, and how to recognize an answer that looks complete without actually solving the problem.
The goal is not a fixed prompt library, but a more reliable way to make human-AI collaboration decisions.