Summary of Key Points
As AI models (such as Claude Opus 5 and Fable 5) continue to improve rapidly, many of the previously popular complex prompts and Skills (the “work manuals” for AI) have become outdated. The key changes now are as follows: Prompts should no longer follow a rigid, role-playing format but should clearly communicate the requirements in a way that resembles human communication; Skills are not necessarily the more the better—instead, they should focus on high-frequency, stable tasks, leverage unique expertise, and have clear acceptance criteria.
Detailed Explanation
1. Prompts: Stop Making AI Pretend to Be Experts; Clear Communication Is Crucial
In the past, when writing prompts for AI, people liked to assign roles to the AI, such as “You are a new media editor with ten years of experience…” However, these predefined roles are not very useful anymore since the models themselves understand what editors do. The focus of prompts should now be on clear communication—just like when you assign tasks to a colleague, you need to specify “what needs to be done, how it should be done, and what the desired outcome is.” For example, if asking AI to summarize a paper, don’t just say “Summarize this paper”; instead, state “Tell me three key points: What problem does it solve? What methods were used? How does it differ from previous summaries?” This will help AI provide more relevant and concise answers.
2. Rules: Less Is More; Avoid Overcomplicating Things
Previously, longer and more detailed prompts were considered more professional. However, too many rules can be counterproductive, leading to AI struggling to prioritize tasks effectively. New models have their own reasoning abilities, so only the most essential rules need to be provided. For instance, if you dislike using single-word verbs (like “do” or “look”), simply state “Use two-word verbs instead, such as ‘complete’ or ‘observe,’” and let AI handle the rest on its own.
3. Skills: Not a One-Size-Fits-All Solution
Skills serve as “on-demand reference manuals” for AI, but not everything can be turned into a Skill. For a Skill to be useful, it must meet three criteria: 1) It should occur frequently (e.g., organizing audio transcripts); 2) The process should be stable; 3) It should involve unique expertise (e.g., adding a human touch to the organization of content by providing data for each point). Skills should also be specific and concise—don’t create a single “writing Skill” that covers everything; instead, break it down into smaller, more manageable tasks like “title generation” or “fact verification.” The main instructions should be brief, acting as a guide for AI, indicating when to use the Skill, what the core principles are, and where to find additional details.
4. Examples and Acceptance Criteria: Don’t Limit AI’s Creativity
While examples can be helpful, they can also constrain AI’s creativity if used improperly. When asking AI to write an introduction, provide ideas rather than a fixed template (e.g., “Use a story, unusual data, or your own personal insights”). It’s more effective to give guidance on the thought process. More importantly, define clear acceptance criteria for the tasks completed by AI. For example, a paper summary should answer questions like “What are the changes? Why did they occur? What impact do they have on ordinary people?” A product analysis should help the team make decisions, not just list features. With clear criteria, AI knows what to aim for.
The essence of these changes is that AI is becoming increasingly intelligent. Instead of treating it like a child in need of constant guidance, we should communicate with it in a way that reflects how we would talk to a capable professional—by clearly stating our needs and providing the right direction, allowing AI to utilize its full potential.