Stop pretending to Your AI: Four prompts for real technical work

You asked for AI prompts, and you got them. Surprise: they're not "10x faster" hacks. They're designed to make your brain hurt. This article gives you four prompts that force you to stop pretending.

You asked for this.

After my last article on “Cognitive Honesty,” — AI Demands We Stop Pretending: Why Knowledge Work Needs a Complete Reboot — many of you reached out. You said, “Fernando, this is a great framework, but what does it actually look like? Show us the specific prompts.”

So, here they are.

But I’ll give you a warning: these probably aren’t the prompts you want. You were likely hoping for clever “hacks” to read 10x faster or make your AI do the work for you.

These aren’t those. These are the prompts you need.

The prompts you’ve been using are a trap. You ask ChatGPT to “summarize this chapter” or “explain this concept simply,” and you get a clean, plausible-sounding answer. You copy it, nod, and feel productive. You’ve just engaged in the 2025 version of highlighting a textbook—a performative ritual that looks like learning but actively avoids the cognitive struggle that defines it.

If your AI use is about minimizing effort, you’re just perfecting your ability to perform.

The real work—the hard, honest, transformative work—requires weaponizing AI to make thinking unavoidable. It means swapping your lazy prompts for “cognitive forcing functions”—prompts that demand you do the heavy lifting.

You asked me to show you the tools. Here they are. Have fun!

1. The “Metabolic Triage” Prompt

The Goal: To identify the 20% of a book that demands struggle and explicitly ignore the 80% that doesn’t.

When to Use It: Before you start reading a new, dense technical book.

Prompt Example:

I am about to read “Designing Data-Intensive Applications” by Martin Kleppmann. My only goal is to understand the practical difference between Raft and Paxos for implementing a consensus algorithm.

1. Which specific chapters or sections cover the core logic, trade-offs, and failure models for these?

2. Which 80% of the book (e.g., replication, partitioning, batch processing, stream processing) can I completely ignore to save my mental energy for this one topic?

3. Is there a single diagram or thought experiment in the book that is considered the “key” to this topic?

2. The “Cognitive Forcing Function” Prompt

The Goal: To force your brain to do the work by teaching a skeptical AI, rather than having the AI teach you.

When to Use It: When you think you understand a new, complex concept.

Prompt Example:

Your Role: You are a skeptical Senior Staff Engineer. I am a new developer.

My Task: I’m going to explain my understanding of the “Singleton Pattern” in software design.

Your Rules:

1. DO NOT tell me if I’m right or wrong.

2. DO NOT give me the correct answer or explain the concept yourself.

3. DO ask me probing, Socratic questions that ruthlessly expose any flaws, gaps, or fuzzy thinking in my explanation.

4. Focus your questions on edge cases, thread-safety, and why I might not want to use it.

I’ll start. Here is my explanation: [Insert your 1-2 sentence explanation here, e.g., “It’s just a global variable that you can only have one of.”]

3. The “Code-First” Triage Prompt

The Goal: To use AI to bridge the gap when the book’s code is confusing and its prose is unhelpful.

When to Use It: When you’re staring at a code block and have no idea why the author wrote it that way.

Prompt Example:

I’m stuck reading “Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow.”

Here is the code:

Python

[Paste the confusing code block here]

Here is the book’s explanation:

[Paste the vague 1-2 sentence explanation from the book, e.g., “We then use a StandardScaler to normalize the dataset.”]

My Problem: The explanation is useless. I ran the code without the StandardScaler and the model still works, it’s just a bit less accurate.

Your Task: Don’t just tell me what StandardScaler does. Instead, what critical question am I failing to ask? Am I missing a core principle of ML (like gradient descent optimization) that makes this step non-negotiable in a real-world scenario?

4. The “Proof-of-Work” Prompt

The Goal: To get a novel problem from the AI that proves you can apply what you just learned.

When to Use It: The moment you finish a chapter, before you can forget the concept.

Prompt Example:

I just finished the chapter on Terraform Modules in “Terraform: Up & Running.” I understand the book’s example of building a reusable VPC module.

Your Task:

1. Give me a new, mini-project prompt that forces me to build a different reusable module.

2. The prompt must not be the same as the book’s example.

3. It must require me to use variable blocks, output blocks, and a README.md.

Once you give me the prompt, I will write the solution. Then, you will act as a peer reviewer and critique my module’s structure, clarity, and efficiency.

That’s it. Four prompts for you to dwell on. Let me know your thoughts!

Peace. Stay curious! End of transmission.