**TL;DR:** AI is a generative tool with "strong interaction + encyclopedia + search engine" characteristics. Its ceiling is heavily dependent on the user's domain experience. In the "Vibe Coding" era, AI can significantly reduce the cost of trial and error, but human creativity and cross-domain thinking remain the core keys to solving complex, unknown problems.

My AI Tool Stack and Scenarios

In my daily work and study, I have established a relatively fixed AI tool chain:

• Common Tools: Gemini > Cursor > Comet > Claude (ranked by frequency).
• Core Scenarios: Writing code, knowledge query/consultation, translating financial news, and summarizing long videos or books.

The Capability Boundaries of Generative AI

AI is essentially a generative tool based on existing data. Its capabilities are limited to content already available online or offline; it cannot directly solve problems it has never encountered.

A crucial reminder: AI is extremely poorly suited for numerical calculations. It might tell you “1+1=3” with a very deceptive string. Therefore, whenever mathematical calculations are involved, be sure to use a programming language (like Python) to have the AI write a script to calculate, rather than letting the LLM give a direct conclusion.

Hallucinations and the “AI Colleague”

I tend to see AI as a personified tool, or a colleague working together. But this “colleague” has strong “hallucinations,” sometimes even claiming a task is done while I watch the progress remain stationary.

In collaborative work, increasing the number of people does not necessarily bring a 1+1>2 effect. Due to communication friction, more people often lead to lower efficiency. Using AI is the same; if you cannot guide it clearly and accurately, simply introducing AI will not bring a revolutionary change to work efficiency.

Transitioning from Traditional Development to Vibe Coding

As early as 2025, I had the idea of making a personal website. At that time, following the traditional method: searching for tutorials, buying domains, renting servers, learning frameworks (Tailwind, React, Vue.js)… the huge time cost and aesthetic challenges eventually made me give up.

By 2026, with the rise of Vibe Coding, I tried developing websites and game automation scripts again. This time, with the assistance of AI, the speed was qualitatively improved, whether it was configuring parameters or front-end/back-end development.

The Ceiling of AI Tools: Dependent on the User’s Experience

I believe the ceiling of AI tools is heavily dependent on the user:

• Experts using AI: Can lead the AI to solve problems quickly and accurately, and can correct hallucinations in time.
• Novices using AI: Are very likely to be led astray by hallucinations without realizing it. Even if the task is completed, the whole process is a “black box” to the user, who has no idea what happened.

Therefore, the emergence of AI will not easily impact the moats of software companies. Senior developers can use AI tools to eliminate repetitive labor and focus their energy on innovation and architecture that AI cannot touch.

Summary: Creativity is the Core of the Future

Ultimately, AI is a tool without original creative ability. It is good at smoothing out information gaps and solving problems in mature fields, but in the face of some emerging fields or complex requirements, it still requires human understanding, cross-domain thinking, and the ability to switch mindsets quickly.

Just as when I was developing game scripts—when the PC side hit a bottleneck, I quickly shifted my focus to the mobile side. This ability to guide AI is the core competitiveness of the future.

Usage Tip: When encountering an LLM that cannot solve a problem after multiple attempts, don’t keep trying; decisively switch to another LLM.