After Clawdbot became a hit, I couldn't resist: A liberal arts student with zero experience practicing Vibe Coding.

Authors: Denise I Biteye Content Team, Bbo I XHunt Operations Team

I recently saw on X that everyone is using Clawdbot to do all sorts of things with AI agents, and I suddenly realized that liberal arts students can also play with code in the "vibe" way, so I started this practical project for beginners.

Vibe Coding is seen as a new paradigm of "requirements as code" and is even expected to "enable non-programmers to build tools".

However, a real question remains: when the technical barriers are lowered, will the ability to truly implement the technology also follow suit?

To answer this question, I experienced three mainstream Vibe Coding tools as a user with absolutely no coding experience.

The process was not smooth: I encountered pitfalls, setbacks, and had to start over repeatedly. But it was precisely because of this that I figured out where the problems were most likely to occur and how to avoid them.

This article is not a tool evaluation, but a true record of cognition and practice.

I. Conceptual Understanding: What is Vibe coding?

This is a new development model that "only requires you to provide requirements, not write code".

This concept was first proposed by Andrej Karpathy, former co-founder of OpenAI, who said:

"This is a new way of programming that I call Vibe Coding. You completely surrender to your senses (Vibes), embrace exponential growth in efficiency, and even forget the existence of code."

In Vibe Coding's model, you are no longer a "construction worker," but more like a "product manager" or even a "client."

Your task is to tell the AI ​​precisely: "I want to build a house, with floor-to-ceiling windows here and a swimming pool there."

If you're not satisfied, just say, "It doesn't feel right, let me change it again."

This is Vibe!

II. Avoiding Pitfalls: The Safety Red Lines of Vibe Coding

Before venturing into AI programming, you must first secure your "vault." Many beginners unknowingly send sensitive information to AI when conversing with it, which is extremely dangerous.

1. Three types of data that must never be disclosed

• API Key: This is like a "prepaid card" for various platforms. If it is leaked, the credit limit will be used up instantly.

• Private key/mnemonic phrase: Absolutely, never, ever send it to any AI or write it in any code file. Once leaked, your on-chain assets will be worthless.

• Cookie Some tutorials teach you to automate the process of obtaining cookies from your browser, but cookies contain your login state. If leaked, others can log in to your Twitter, Discord, or even exchange accounts without a password.

2. Core Principles: Environmental isolation, data anonymization

Never write API keys or sensitive information directly in code or hints.

The correct approach is to use "environment variables" so that the code reads the system's variable names instead of directly reading the key content.

The configuration steps for the Windows system are as follows:

• Press Win+S on your keyboard, type "environment variables" in the search bar.

• Create new variable

• Enter variable name and variable value and save.

The configuration steps for macOS are as follows:

• Press Command + Space on your keyboard to search for and open the Terminal.

• In the pop-up black box, copy and paste the following command (be sure to replace it with your real key), and then press Enter:

  echo 'export CRYPTOHUNT_API_KEY="your sk-xxxxxx key"' >> ~/.zshrc

• To apply the configuration immediately, type and press Enter: source ~/.zshrc

• (Optional) To verify success, enter: echo $CRYPTOHUNT_API_KEY. If your key is printed on the screen, the configuration is successful.

III. Real-world (or rather, unfortunate) debriefing: Three tools, three experiences

Now that the theory is covered, let's move on to practical application. To find the most suitable tools, I've tested three mainstream tools for you. The reason for choosing them is that they represent three mainstream forms of current Vibe Coding:

• Google AI Studio: Represents a lightweight approach to "rapidly generating prototypes".

• Antigravity: Represents the engineering roadmap of "AI agent + local full-stack development".

• Lovable: Represents the abstract approach of "dialogue as application".

If you only want to know which tool is best suited for beginners to implement directly, you can skip to the third tool, Lovable; if you want to avoid detours, it is recommended to read this section in its entirety. Let's begin the review in order.

1. Google AI Studio

Target audience: Suitable for individual developers and startup teams

Experience: Emphasizing "speed" and "free," it allows you to validate an AI idea in minutes.

Official website: https://aistudio.google.com/apps

I first used it at Christmas. Everyone was making cyberpunk Christmas trees, so I gave it a try. I casually threw out a prompt, and in just 5 minutes, it generated a very beautiful 3D page. The particles flowed with the gestures, and the effect was extremely smooth.

This makes me realize how much times have changed.

(Think back to how, in the past, a tech guy would stay up all night writing a bunch of code to make a girl happy, and after running it for ages, a heart would pop up – and that was considered pretty cool. But now? You can just leave it to Vibe Coding, and it can deliver maximum emotional value in no time.)

Emboldened by the success story, I started to get "overconfident" and began to do something serious: using the Cryptohunt API to create a Twitter data dashboard.

Full of confidence, I fed it the API documentation. Everything went smoothly at first; the backend logic worked instantly, and the data was successfully retrieved. I thought to myself: Is that all?

However, when I asked it to "present the data beautifully on the front end," the nightmare began:

• The chart isn't displaying? It needs code modification.

• It's displayed but not fully? They changed the code again, and now the interaction is broken.

• Is the interaction fixed? Now the API data isn't loading again.

I found myself in an endless cycle of "fixing bugs—creating new bugs." After two days, I gave up.

Final image (the AI ​​has started drawing random symbols on it).

Summary of my experience: Google AI Studio is a jack-of-all-trades, master of none. It's amazing for creative projects and single-page demos. But when it comes to deep front-end and back-end integration (like API data flow and complex front-end displays), it easily falls short. If you have no understanding of code logic, you can easily get stuck in a dead end with the front-end presentation, just like I did.

2. Antigravity

Positioning: A proxy-based full-stack development tool integrating multi-model switching.

Features: Multi-model switching, task breakdown, automated testing

Official website: https://antigravity.google/

In actual use, I almost gave up before I even started writing any code because of its login feature. Unlike most tools that you can start using right away by opening a webpage, it has rather strict requirements for account and network environment.

Account restrictions are not determined by your current network IP address, but by the "country/region associated" of your Google account. If your region setting is not on the supported list (e.g., Mainland China, Hong Kong), your login will be rejected by the system. My requests to change my region have been rejected several times.

After going through the whole process, I can only describe it in one sentence: I was completely numb.

After having three accounts banned and switching environments multiple times, I found a way to log in to the "evil cultivation" site: Antigravity tools.

1. At the start, you need to set up a US IP address and use the global Tun mode.

2. Download antigravity tools

   Link shared by a Twitter user: https://x.com/idoubicc/status/2004848130693759213

3. After the download is complete, add your account and begin OAuth authorization.

4) Once authorization is successful, you will be redirected to the login page, where you can also see your model quota.

(Risk Warning: Antigravity Tools is an open-source project that facilitates the integration of Antigravity models with tools such as Claude Code. However, during Vibe Coding, it is recommended to isolate the operating environment from the computers used for asset wallets or important accounts to reduce potential security risks.)

After getting inside the tool, I also summarized two extremely important tips for beginners:

1. Install the Chinese language pack: Search for "Chinese" in the left-hand extension bar, install the Simplified Chinese language pack, and restart the software. This step will change the interface to Chinese, significantly reducing the learning curve.

2. Inject a "Global Persona": Antigravity allows you to set a "global rule" that the AI ​​must always follow. How to do it: Click the Rules icon on the right → Select + Workspace (Add Workspace Rule) and paste a "global rule hint." This forces it to use Chinese and significantly improves code output quality.

However, even after successfully logging in and completing the basic configuration, I was still met with a harsh reality when I actually started running the code. It didn't offer an instant preview like Google AI Studio. I had to download a bunch of accompanying preview software, such as Node.js and VS Code, or simply view the results in HTML. Moreover, the response time for large models was visibly slow. Every time I submitted a requirement, the screen would just spin.

But Antigravity has a truly amazing feature: automated testing. After you finish writing the code, it will launch a video preview and, just like a real person, click around on web pages to check if the logic works. Watching the AI ​​click around on the screen itself gives you a real sense of awe that "the future is here."

The project I had it try to do was: "A cryptocurrency alpha signal catcher. It uses the Cryptohunt API to monitor Twitter data, automatically grabs ticks from tweets, counts which tokens people are talking about, and sorts them by popularity."

Reality is harsh. Although it seemed to be working hard to help me write code and test, I still ran into trouble at the core step of API integration. It might be due to the complexity of the local network environment, or it might be due to the AI's misunderstanding of the API documentation; the data just wouldn't connect.

Demo: Display effect of simulated data

Seeing the red error messages in the terminal and the AI's illusion that "this time it will definitely work," I realized: while full-stack development is great, due to the black-box nature of networks and environments, creating a truly usable tool still has an extremely high barrier to entry for beginners. This is because Antigravity assumes you're already a developer.

3. Lovable

Positioning: "Conversation as Application" platform

Features: No local environment configuration required, WYSIWYG (What You See Is What You Get)

Official website: https://lovable.dev/

Using Lovable made me realize what a real client is like. Is it because they're the only ones who have to pay? (Tip: I spent 15 RMB on 100 credits on Xianyu, and I haven't used them all yet after finishing the project.)

On Lovable, without changing a single line of code, I successfully set up the following Crypto Twitter monitoring tool—Crypto Pulse—as a monitoring dashboard.

Yes, you read that right, I finally succeeded this time.

Online trial available here: https://tweet-whisperer-dash.lovable.app/

"Function Details"

A. Summary of Panoramic Data and AI

Upon entering the homepage, the first thing you see is a "thermometer" of content:

• Real-time data cards: The top displays the total number of tweets, total interactions, number of active KOLs, and market sentiment index within the past 24 hours. Without even looking at the specific details, and by comparing the data with the previous 24 hours, a glance at whether it's red or green tells you whether the market is experiencing fear or greed today.

• AI Smart Summary: This is the most time-saving feature. The system integrates Gemini AI, which automatically reads thousands of tweets from the past few hours and then writes a market summary for you.

B. Trend Radar

Don't want to miss out on the next 100x coin?

• Intelligent categorization: Automatically filters trending topics and categorizes them into Topic, Project, Token, and KOL.

• Market Trend Indicator: Next to each tag is the 24-hour mention rate change (increasing/decreasing). Where discussion surges, there is opportunity.

C. Intelligent Information Flow

Say goodbye to the messy timeline, here are four cleaned-up tabs:

• Highlights: High-value content that has been secondarily filtered by AI, eliminating meaningless spam.

• Trending: The most popular tweets with the highest engagement (likes + shares).

• Influencers: Only view real-time posts from industry leaders (such as Elon Musk, Vitalik, etc., with more than 100,000 followers).

• Latest: Stay updated with the latest tweets from CT-related accounts.

  *User experience details: Supports filtering by region (English-speaking/Chinese-speaking regions), and clicking on the card will directly jump to the original Twitter post.

"Usage Reminder" Because I am currently using my personal API Key, data return will consume API Credits, and the quota is limited. I suggest you try it out as soon as possible.

IV. Cryptohunt API Practical Guide

During my coding process on Vibe, I chose the Cryptohunt database, which offers good value for money. While I didn't master the coding process, I did thoroughly understand the API, and I'll write a guide about it here.

Official link: https://pro.cryptohunt.ai/

1. Precision feeding

Many beginners will simply throw tens of thousands of words of API documentation at the AI, and the AI ​​will be overwhelmed.

• Incorrect example: This is the API documentation, you can figure it out yourself.

• Correct example: I want to implement a token monitoring feature. Please read the section on post/tweet/mention_tweets in the Cryptohunt API documentation. Tell me how to get the data for the most recent hour?

Example prompt: "Please call the /tweet/mention_tweets API to search for tweets with the ticker 'SOL'. The time limit is the most recent 24 hours. Please filter out the top 5 tweets with the highest engagement (likes + retweets) to see how people feel about SOL."

2. Debugging tool

The most frustrating situation while coding on Vibe is when you click a button but nothing happens on the screen. At this point, beginners often start to doubt themselves: Is the AI ​​code wrong? Is the network slow? Or is the API down?

This allows you to enable API usage refresh:

• The record shows points were deducted: This indicates the API call was successful! The data has been successfully sent back to your computer. This is a front-end display issue. You can tell the AI: "The data has been received, but it's not displayed on the screen. Please check the rendering code."

• No record at all: This means the request wasn't even sent! So it's a logic/network issue. Tell the AI: "Clicking the button has no effect; the request wasn't sent. Please check the click event."

Simply feed the conclusions to the AI, and it will fix the problem much faster. In the process of Vibe Coding, the real energy drain is no longer on writing code, but on determining where the problem lies.

Once we delegate the complex syntax and implementation details to AI, the focus of development shifts back to controlling information.

Therefore, people often say that the ultimate goal of Vibe Coding is not code, but data.

V. In conclusion: Master the two core techniques of Vibe Coding

Having gone through the "trial run" with Google AI Studio, the "torture" of Antigravity, and then the "explosion" of Lovable, I finally understand: Tools are just swords in your hand; your Vibe is the path in your heart.

When trying Vibe Coding, please remember these two core points; they are more important than any code:

1. Dare to dream: Imagination is your only "ceiling".

• In the era of Vibe Coding, technological barriers have collapsed. Previously, we hesitated to create tools because we feared we couldn't write the code; now, practical experience has proven that as long as the logic is sound, AI can write it.

• Don't just be a "bricklayer," be an "architect": Dare to conceive of products you never even dared to imagine before. Is it a radar monitoring all network data? Or an automated arbitrage robot? What you need to do is transform that vague "desire" in your mind into a clear "blueprint."

2. It will say: The prompt word is your source code.

Many beginners fail not because the AI ​​is stupid, but because they are too "polite" or too "vague".

Fuzzy demand = fuzzy result

• Reject ambiguity: Don't tell AI, "Make me a user-friendly website." (AI: What does "user-friendly" mean?)

• Embrace precision: Say things like I would when creating a dashboard: "I need a dark mode dashboard with these four data cards at the top, this API in the middle, and AI-generated summaries on the right..."

Remember, you are now the client. The client's authority comes from clearly defined needs and patient feedback. If the AI ​​makes a mistake, don't doubt yourself; just tell it, "Wrong, start over. I want A, not B."

If you've read this far, you've likely already developed the ability to create tools using Vibe Coding.

Don't hesitate—

Quickly turn your Vibe into Coding.