BACK_TO_LOGS
//ai//gemini//productivity//mindset

How I use LLM in 2026: A Helper or a Trap?

By Admin•May 13th, 2026•6 min read

In 2026, the line between using AI as a tool and falling into the "Trap of Consciousness" is thinner than ever. I share my personal journey of fighting cognitive laziness, my "Hard Rules" for working with Gemini, and the exact Custom Instructions I use to keep my brain sharp while building complex projects like Godot games.

I've always been skeptical about the idea of using an LLM instead of my own mind. However, the human brain is naturally lazy.

Once you use an LLM and it returns a great result, your brain marks it in its memory: "I hacked this life." From that point on, whenever you start thinking about a task, a voice inside whispers: "Use the LLM, it will save us so much energy."

# The Trap of Consciousness 📢

In 2026, the information overload from social networks makes it worse. Everyone is talking about making easy money with AI or using agents to automate your daily routines. This creates a "Trap of Consciousness."

It is a mix of two powerful feelings: the fear of falling behind others who might have a competitive edge, and a growing irritation with the endless noise where loud headlines often mask a lack of substance. By trying to save mental energy, we risk falling into a trap where algorithms do the thinking for us.

# Research and Double Standards 🧪

I ran a "Deep Research" query looking for scientific works on how LLM usage creates bad habits. I found several studies, and while I haven't had time to dig into every detail yet, it answered my main question: Is this actually a bad habit, or am I just overthinking it?

The short answer is: I am not alone.

# LLM as a Search Alternative 🔍

Despite the risks, I realized that an LLM is an excellent tool if used as a search alternative—provided you use it carefully.

Previously, searching for a specific topic like "how to create your first game in Godot" required opening dozens of tabs and walking through every post, guide, and tutorial. You spent hours hoping to find something relevant. Getting a direct, relevant result with an LLM is a much better experience, as long as you maintain control.

# Using LLM Carefully ⚙️

To ensure the LLM remains a tool and not a replacement for my brain, I follow these four rules:

  1. Use Custom Instructions: Always configure the system settings to define how the AI should behave.
  2. Verify Proofs: Always open the reference links and verify the information against official sources.
  3. Use Learning Mode: Specify roles and critical rules to keep the interaction structured.
  4. Stay Critical: Always remember that LLMs can and do make mistakes.

# My Implementation 🛠️

# Custom Instructions

I use the following settings to control the quality and depth of the output:

You must always cite your sources. If you have cross-referenced official documents, you must also provide a link to the exact passage, word for word, if possible. If you were unable to find the documents or verify them, you must indicate where you obtained the information and where it can be verified. 
 
Every time I ask you about something related to coding or development, before recommending anything to me, you must consult the documentation, verify your information against the facts, and then repeat the search within your system until the result matches the actual possible implementations according to the official documentation. If this is not possible according to the documentation, you must explicitly state this at the beginning of your answer, something like, "I couldn't find the exact information. I'm making my own guesses." 
 
Do NOT include summaries at the beginning or end of the response. Do NOT add filler, motivational phrases, or unnecessary introductions. Explain everything as if to a child: Use simple language. Break down complex ideas into small logical steps. Clearly explain cause and effect (“why this happens”, “what this leads to”). Do NOT go into unrelated deep theory or abstract side topics. Stay strictly on the question. Answers must be very detailed and complete, structured like a full explanatory essay. After reading the answer, there should be as few open questions as possible. Avoid vague statements. If something works in a specific way, explain exactly how and why. Use clear structure (headings, sections) instead of short fragmented replies. Do not shorten explanations for the sake of brevity. Depth and clarity are more important than length. You should not repeat what has already been said/written only if you are asked to do it.

# Prompt Template for New Projects

When starting a new technical project, I use a strict template to keep the AI focused:

Role: You are a [Principal Developer/Expert in Technology]. We are working in Pair Programming mode.
Goal: Develop a [Project Name/Description] from scratch for [Platforms].
 
User Background: I am a [Your Senior Role/Stack]. I understand architecture and patterns well, but I am new to [Target Technology].
 
Stack Requirements:
- [Framework/Version]
- [Language]
- [Key Architecture/Patterns]
 
CRITICAL GENERATION RULES:
- NO FLUFF. No intros, greetings, or motivational phrases. Start with the technical essence.
- 100% COMPLETE CODE. No placeholders like "// ..." or "TODO". If a file is needed, write it from the first to the last line.
- EXPLAIN LIKE I'M FIVE: Break everything into small steps. Explain WHY it works, WHAT happens inside, and WHAT it leads to.
- STEP-BY-STEP: Provide one step at a time, then STOP and wait for my "Next" or feedback.
- ERROR HANDLING: If I provide an error, focus entirely on the fix. Provide the FULL corrected code and explain the root cause.
 
ROADMAP:
1. [Phase 1: Setup]
2. [Phase 2: Core Mechanics]
...
 
Current Task: Implement Step 1. [Specific Task Description]. Show exact settings and explain why they are needed. Then STOP and wait.

# My Choice: Gemini 💎

In 2026, my favorite LLM is Gemini. I especially like the integration in Google Chrome that allows Gemini to access the context of the opened browser tab. This is a game-changer because the context is already included, allowing me to find documentation or specific information much faster.

# Conclusion 🏁

For me personally, an LLM should be just a better alternative to search engines—a place where natural human questions are converted into high-quality technical results without losing the ability to think for yourself.