The Problem with Being a Taskmaster
You might be accidentally handcuffing your AI tools by being too specific with your instructions, limiting the model’s intelligence to your own current knowledge level.
I just read a fascinating Reddit post by user AdCold1610 that completely flips the script on standard prompt engineering. The original poster shared a revelation that changed how they used AI after eight months of struggle, shifting from a command-based approach to a consultation-based one. Instead of acting like a boss barking orders, the expert started acting like a colleague seeking advice, and the results were transformative.
The “Consultant” Shift
The core concept here is deceptively simple but incredibly powerful. The author explains that when you tell an AI exactly what to do, it does exactly that, and nothing more. It follows your instructions to the letter, even if your instructions are flawed, incomplete, or based on outdated assumptions.
For example, the LinkedIn user notes that in the “old way,” they would say, “Write a function to validate emails.” The AI would dutifully spit out a block of code using a standard regular expression (regex).
In the “new way,” the author asks, “I need to validate emails. What would you do?”
The difference in output is staggering. Instead of just writing a script, the AI pauses to actually think about the problem. It suggests using regex, sure, but it also recommends checking for disposable domains, validating MX records, and adding a verification step. By asking for an approach rather than a specific task, the creator unlocked a comprehensive strategy that a simple command would have missed completely.
💡 Insight: The “Unknown Unknowns” Problem
One of the biggest risks in prompt engineering is that you don’t know what you don’t know. The post’s author highlights a perfect example involving API performance that illustrates this trap beautifully. In a traditional workflow, you might realize your API is slow and tell the AI, “Write code to cache the API response.” The AI will do exactly that. It fixes the specific symptom you identified using the specific method you requested.
However, when this innovator asked, “What would you do to speed up this API?”, the AI provided a much broader solution. It suggested caching, but it also recommended implementing request debouncing on the client side and using connection pooling on the backend. The author admits they weren’t even thinking about the client side. This is the critical value add: the AI has access to a massive database of engineering best practices. When you give a narrow command, you restrict the AI to your own current knowledge level. When you ask an open-ended question, you allow the AI to tap into its wider expertise and point out blind spots you might have missed.
📌 Insight: Shifting from Tool to Teammate
The mental shift described by the original poster is about moving from a “boss/employee” dynamic to a “peer/peer” dynamic. When you act as the boss, you assume you have the best plan and just need someone to execute it. This works fine for simple tasks, but it fails for complex problem-solving where you might not have the full picture.
The expert suggests that treating the AI like a teammate unlocks better reasoning. By phrasing prompts as consultations, like “How would you approach this?” or “If this was your problem, what’s your solution?”, you encourage the model to simulate a reasoning process. It stops acting like a typewriter and starts acting like a consultant. It explains why it chose a certain path, often educating the user in the process. This takes the pressure off you to be the sole architect of the solution and allows the AI to do the heavy lifting regarding strategy.
✅ Insight: Practical Phrases to Unlock Reasoning
The Reddit contributor provided several specific variations of this prompt that you can start using immediately. The goal is to prompt the AI to evaluate the landscape before it commits to a path. The author notes that this method forces the model to generate a high-level plan before diving into the details, which results in more robust solutions.
Here are the key phrases the author recommends adding to your workflow:
“How would you approach this?”
“What is your move here?”
“If you were building this, how would you structure it?”
These questions often result in “chain of thought” reasoning, where the AI outlines its logic. This is vastly superior to a zero-shot answer where it tries to guess the code immediately. You get the benefit of a second pair of eyes on your architecture, rather than just a pair of hands on the keyboard.
How to Apply This Today
You can test this immediately with your next task. Instead of defining the solution step-by-step, define the problem and ask for the strategy.
Scenario: You want to create a marketing plan.
Old Prompt: “Write a 5-email sequence for a new coffee product.”
New Prompt: “I am launching a new coffee product and need an email sequence. If you were the marketing director, what strategy would you use to ensure high open rates and conversions?”
Scenario: You are debugging code.
Old Prompt: “Fix the syntax error in line 10.”
New Prompt: “I’m getting a syntax error here, but the logic also feels clunky. How would you refactor this block to make it cleaner and bug-free?”
This simple tweak shifts the burden of strategy from you to the machine, often with surprising results! Check the full Reddit thread by AdCold1610 for more discussion on this technique.
💡 FAQ & Troubleshooting
Why is asking “What would you do?” better than giving direct instructions?
When you give specific commands (e.g., “Write this function”), the AI strictly follows orders and often misses broader context. By asking “What would you do?” or “How would you approach this?”, you shift the AI from a tool to a “teammate.” This prompts it to analyze the problem holistically, often suggesting critical improvements—such as caching strategies, edge-case handling, or security checks—that you might not have known to ask for.
How can I ensure the AI writes high-quality, long-term code?
To avoid “janky” implementations or lazy coding, add specific quality constraints to your prompt. Explicitly ask for the “most robust and future-proof way” to accomplish the task and state clearly that you “don’t want shortcuts.” This forces the AI to prioritize best practices over the simplest working solution.
What is the best workflow for building complex features?
For complicated tasks, instruct the AI to adopt specific “hats” or personas in a sequence. Start by telling it to be a Product Manager to create a PRD (Product Requirement Document), then a Software Architect to design the structure, and finally an Engineer to implement the code. Review and refine the documentation at each stage before moving to the next.
Should I let the AI make the final decision on an approach?
No. AI is generally weak at executive decision-making. You should use it to explore “Option A vs. Option B,” list pros and cons, or suggest an “Option C” you hadn’t considered. However, the human user should always make the final decision on which path to take.
I accidentally broke ChatGPT by asking “what would you do?” instead of telling it what to do
byu/AdCold1610 in