Grill Me, the Skill That Cooks Your Brain

grill-me is a popular AI Skill popularised by Matt Pocock. The skill itself is quite simple:

The intention is that you use this skill to squeeze out all details of an idea you have. Whether it is a plan, product feature, or whole application architecture. The AI is just helping you think more, that sounds great, right!?

I think grill-me is generally bad and illustrates a lot that's wrong with hype culture and specifically how it ties in with AI. grill-me effectively outsources your critical thinking to AI, the prime trap of AI. You might think "but grill-me presents me with questions? I need to answer and think about them? Doesn't that promote critical thought?". The question is a carrot on a stick, the AI has supplied you that question, supplied you that line of thought. In grill-me it also often supplies you with multiple choice options and provides a recommended option. This falls directly into the trap of choice architecture, and that's not a technical term, but a psychological one from the field of behavioural economics.

Choice architecturerefers to the deliberate crafting of decision-making environments. By subtly shaping how options are presented, choice architecture influences individual decision-making, often without their explicit awareness.

This encourages you to not think critically about the problem you're facing. The options presented narrow your field of view of the problem space and also the decisions you could possibly make in that space. It makes answering feel like thinking, while quietly moving the hard part of thinking, deciding what matters next, into the AI.

AI will always regress back to its homogenous origins, that is to say, the most common high-probability concepts in its training data. It means you'll be making the same decisions everyone else is, while feeling like you're being productive and being led down the same hole as every other person. The result? You're burning tokens, wasting time, and in the worst case, grill-me becomes auto-spec-generation with extra steps: you click through AI-framed decisions and feel a greater sense of false-ownership .

Can We Fix It?

In its current form, no, not really. Asking AI for feedback or using it as a means of brainstorming itself is not bad. The difference between you asking the AI for feedback and grill-me is that you are driving the conversation. You are in control of where the train of thought is going. You are making the decision. You have to think of "what is the next course of action here?". Versus grill-me, which is a conveyor belt of silver platters, you choose the one to eat and the AI fetches the next four. You make AI answer your questions, not the other way around.

The best way to frame this is:

AI is good at highlighting new angles to a problem you're thinking about.

For instance, "What if this endpoint is hit by one million users?", "What if we toggle feature flags A, B, C, and D?" or problem spaces that are bounded: "Are we GDPR compliant?", "Do we leak any PII?" etc. Essentially any scenario where a check-list style interrogation would help, anything fuzzier than that lands you back in the trap.

A good grill-me skill is literally you, yourself, asking the AI targeted questions that spawn from your own train of thought. Then you react to that feedback to drive your next train of thought. You remain in control and thinking.

Broader AI Hype Culture

grill-me highlights the issue with the broader AI hype culture: no one knows what they're doing, and everyone rides the hype train of popular skills because they're popular, even if the skill is to your detriment or the detriment of your workflow.

So what can you do? I'd suggest the following:

• Experiment: Use other people's workflows, skills, and tools as inspiration and not gospel. Take a skill, extract the bits you take inspiration from, and use them in your own skills or workflows.
• Stay focused and thinking: Resist the temptation to switch your brain off and use skills or workflows that encourage it. If you're not experiencing friction in your workflow, that's a big red flag. AI is here to write the code and do grunt work, not do software engineering.
• Learn: AI is a whole new paradigm. You have to learn it from first principles. Don't shortcut it. Additionally, understand what the AI code does; do not let it code something beyond your capabilities. Use it as a learning resource.

Closing Remarks

There are many other examples of AI trends negatively impacting people's effectiveness with AI. This is but one of them. The landscape is vast, complex, and rife with small details, exactly like learning to code all over again. If you take one message from this article, let it be this: Make your own workflow; don't become someone else's.