The video discusses common misconceptions and mistakes people make when using the “grill me” and “grill with docs” AI skills, which are designed to replace traditional planning modes in engineering projects. The core idea behind these skills is that they relentlessly question the user to reach a shared understanding of a problem or project. However, success with these tools depends heavily on the user’s skill in planning, understanding scope, and knowing which questions require detailed answers. The presenter emphasizes that these AI skills are meant to assist engineers, not replace them, and mastering their use requires active engagement and strategic thinking.
A key concept introduced is the distinction between low-fidelity and high-fidelity questions. Low-fidelity questions can be answered directly in a grilling session, such as deciding on URLs or routes, while high-fidelity questions require detailed prototypes or actual builds to answer, like user interface feel or form design. The presenter advises against trying to answer high-fidelity questions during grilling sessions, suggesting instead to hand off these questions to dedicated prototyping sessions and then return to grilling once more concrete information is available. This approach helps maintain focus and efficiency.
Scope management is another critical factor. If the scope of the grilling session is too large, it can lead to hidden high-fidelity questions that are difficult to answer without concrete examples, and it can also overwhelm the AI model’s context window, pushing it into what the presenter calls the “dumb zone,” where the model’s performance degrades. To avoid this, users should break down large scopes into smaller, manageable parts and grill them individually. This not only keeps the AI effective but also helps maintain alignment and build on solid foundations rather than speculative future plans.
The presenter also highlights the importance of being active rather than passive during grilling sessions. Users should lead the conversation, manage scope, and keep the session on track rather than passively answering endless questions. However, being too pushy and grilling low-fidelity questions endlessly without moving to implementation is also a pitfall. Additionally, preserving the valuable design decisions and context generated during grilling is crucial; users should avoid clearing the context prematurely and instead create handoff documents or PRDs to capture and utilize the session’s output effectively.
Finally, the video stresses the importance of using a capable AI model for grilling, as these sessions rely heavily on the model’s parametric knowledge to generate insightful questions and suggestions. While simpler models may suffice for implementation phases that rely more on contextual knowledge, grilling benefits from more advanced models. The presenter also recommends running multiple grilling sessions in parallel to increase throughput and efficiency, comparing it to managing multiple Slack threads. Mastery of these skills can significantly enhance planning and engineering workflows, and the presenter invites viewers to join their AI coding cohort for further learning.
