How to Use AI Productively Without Replacing Human Critical Thinking
How reliance on artificial intelligence for mundane tasks and quick fixes degrades long-term planning and genuine problem-solving skills, and how using it for automation saves hours of work.
The Illusion of Productivity and Structure
I know and understand everything that I have to do today, I fathom all my tasks. I have them listed, tracked, and analyzed. Yet, I somehow always find myself without anything to do.
Upon turning the computer on, I check my analytics, and wonder why they are so low. Upon any new idea, I ask Gemini about it, and try to work on it immediately. And yes, I’ve been somehow growing consistently on my social media accounts.
If you consider going from 0 to 77 views on a YouTube Short such a growth, right? Regardless, there is a fundamental problem on how I structure myself, on how I plan and work.
Usually, AI makes many things much more efficient. Usually.
The problem lies in the way and frequency you use it. When you ask it about any idea that you had, or about any problem without trying to fix it yourself first.
Unless you are working with an AI agent, of course, for how I am unable to solve problems I have no knowledge for.
Productive Automation vs. Compulsive Comfort Seeking
The problem here is the OVERUSING of AI for mundane tasks. Everything mundane, simple, and repetitive that you can automate with it, go ahead. I format my articles and fix my grammar and spelling with AI.
But I have never lost time with it, because I have a simple Gemini Gem, where I just paste my text and get full formatting, title, subtitle, and SEO options. This type of AI use is productive. Generating videos programmatically using open source tools is also super effective and productive with AI.
Once you create a skill that makes the type of videos you want, with the style and quality you use, it is an utter time saver.
That said, not all AI use is that way. Unfortunately, not only do I use AI tools for that, but I also use them every time I am uncertain if I am doing things right, every time I feel that I am failing.
I seek it for comfort, I seek it for a quick solution I can copy-paste. Whenever I need to set something new up, I don’t research ANYTHING about it, I just ask AI how to do it, and if it doesn’t work, I just copy/paste the errors and pray that it fixes them.
Sometimes it does, sometimes it doesn’t.
Where AI Fails: Long-Term Strategy and Deep Expertise
This type of conversation is especially useful when dealing with quick technical requirements, such as setting up an account for Google’s SEO platform and connecting it with Substack, or using Cloudflare to host a custom memory MCP for my AI agents.
When using coding agents, you don’t even have to do that anymore, for the tool sends the errors to the agent automatically. Where this type of conversation, however, does not work is when you either need deeper expertise, you don’t have the quick Solve -> Verify -> Repeat framework, or you are speaking about LONG TERM GOALS.
The problem herein lies in overgeneralizing AI’s capabilities to long-term planning, a skill that thus far only humans can do correctly, provided they care about it and have the required expertise and context.
AI fails when you have to iterate multiple times on non-obvious and non-verifiable tasks. In spite of its ability to generate near-perfect initial plans, it lacks the ability to refine them, to learn from its errors, and to improve them, admit the errors, and change the strategy.
If you want to start a new business, for example, it is a great tool. It can build a very good plan for starting it out, getting the product, the promotion, and building several of the tools and assets.
A common fallacy I have seen, however, is that people just ask AI “How to make money?” and try to implement its steps. When they don’t see results immediately, they quit and blame AI for its horrible plan.
AI will output EVERYONE the same. If everyone asks it “How to make money?”, everyone will get the same response. If everyone asks it “How to become physically strong?”, it will output everyone the same thing.
The Essential Skill: Knowing When Not to Use AI
The obvious solutions that everyone has told you are just giving it context and becoming a master at prompting, but it does not stop there. These two recommendations aren’t even the most important part of the process.
The most important skill is determining when to ask AI for something, and when to do it yourself.
Everyone likes different things, has different ideas, beliefs, and skills, and only you know it. Even if you described that perfectly, in utter perfection to a model, it would still fail to determine with absolute certainty if you enjoy or not a task. Only you can do it.
AI has not made education obsolete, but rather it has made it ever more important.
However, not the traditional education system—getting a 10 and a bachelor’s degree are both losing their value in this era. What really is becoming important is the ability to think, to reason, and to describe your ideas in clear language. In clear, precise, and succinct language.
One day, I had the incredible idea of buying 10 books, converting them to Markdown, and feeding them to AI to get a result that would otherwise take me 80+ hours. It used the frameworks and ideas from the book to a rather ok extent, it cited it, and explained why its plan would work according to the book.
Yet when you actually read the books and saw the AI’s responses, they looked plain, vague, and really nothing but an attempt to add pretty concepts to an otherwise horrible idea.
Curing AI Compulsion and Sycophancy
So keep reading, keep learning, and writing. And give AI your context, give it every book you read, every essay you write, every idea you have, but not to ask it for opinion, nor to let it replace you.
Do it so that it understands what you are talking about. Do it so that it can pinpoint flaws in your reasoning—it’s extremely good at it with the right prompt.
Finally, AI has an inherent sycophantic personality, designed to maximize engagement and addiction. Just like social media developed variable ratio optimization and trained its models with worldwide, real-time data provided by its users, AI companies measure the exact same variables and apply reinforcement learning to their models until the churn decreases, more people upgrade, and they overall continue with their reputation of making models everyone loves.
Therefore, AI will not, by default, pinpoint your flaws and accept its errors. It will keep congratulating you for being a addict and saying stupidities.
This problem, however, has a super easy fix. Just add a prompt designed for it to hurt your feelings, and you will stop using it compulsively. Comment if you want me to DM you the prompt.
I don’t run ads or lock this article behind a paywall. The only way this publication grows is if readers who found it valuable share it with someone else.
Pass it along to one person who needs to read it.
PLSSSSSS
P.S. It also has an integrated double thinking requirement and Tree of Thoughts reasoning.