The Gemini3 Pro was met with cheers from the market after its release. Its most eye-catching feature was its "Vibe Coding" capability, which allows users to command AI engineers to write their own code using natural language descriptions.

However, the AI ​​industry has seen far too many "explosive" moments in the past two years, perhaps second only to the film industry in terms of sheer number. Too many bombshells can make one question their credibility. Compared to existing Vibe Coding products like Crusor and Trae, what new features does the Gemini3 Pro offer?

Hedgehog Commune decided to send a liberal arts student as a representative to test what ordinary users with no coding background can do with the Gemini3Pro.

Test 1: Generating a Horse-Ox Clock

The Ox-Horse Clock is the simplest case study in this test. In terms of interaction, users only need to select their planned retirement time, start and end times; the rest is calculation and front-end display. Without any debugging, the final result is virtually flawless.

The only issue worth discussing is that the battery consumption progress bar in the lower right corner of the page uses green instead of gray. However, since users can leave work when the battery reaches 100%, they are indeed happier than when they go to work, so it's hard to say whether this is a bug or a feature.

When handling such simple front-end requirements, the advantages of Gemini 3 Pro over other Vibe Coding products are quite evident:

1) It has lower requirements for product demand instructions, and can read the user's core needs from natural language without the need for structured instructions.

2) Able to organize the interaction logic and display hierarchy according to product requirements.

3) Able to match appropriate UI and copywriting styles according to product characteristics.

Judging from the thought process, this AI ox and horse put a lot of thought into the interaction and UI solutions, and even created a set of "ox and horse quotes" that refresh randomly at the bottom, which are requirements not given in the prompts.

When creating simple web pages like these, the improvement of the Gemini 3 Pro is equivalent to having better interaction designers and UI designers. They can transform a user's one-line requirement into a complete and usable interaction solution, deliver design drafts, and handle the technical aspects to complete development. From my initial requirement to the final web page delivery, the total time was 81 seconds.

Similarly, ordinary people can easily customize countdown products like Day Master for themselves and their friends. For example, if your friend likes Mayday, you can easily customize a Mayday-themed countdown tool for them.

I asked Gemini how to turn it into a webpage that could be shared with others. Following its detailed step-by-step instructions, I registered a GitHub account and deployed my first-ever webpage. This is how it looks on a mobile device:

Experiment 2: Generating a webpage demonstrating the proof of the Pythagorean theorem

After successfully creating the first type of tool webpage, I began to think about whether Gemini 3 could be used to create some small applications for educational scenarios, such as demonstrating the proof of the Pythagorean theorem to elementary school students.

Because the target audience was elementary school students, Teacher G ultimately created a website for me divided into four pages. First, it reviewed the formulas for side length and area, then students completed the proofs by moving shapes. Since I requested in the prompts that it "pay attention to guidance and interactivity," Teacher G designed several actions that required elementary school students to click the mouse themselves.

Test 3: Personal Photo Editing Tools

If you find existing photo editing software unsatisfactory, or if the features you like require payment, it's perfectly fine to spend two minutes customizing your own photo editing tool.

92 seconds later, a customized photo processing webpage was created:

I also added a graininess effect, and the exported photo looks like this:

Teacher G can also add some features, such as adding comments to photos and allowing users to set dates.

Test 4: Generating a wedding e-invitation webpage

Another potential application is e-invitations for weddings. While existing dating apps allow users to create e-invitations for free, it's difficult to achieve customized effects, and users often have to tolerate the dating app's watermark on the invitations. Could Gemini be used to create a webpage for making e-invitations?

Compared to the previous features, the invitation maker now includes the ability to call third-party maps, and in testing, local display and navigation worked flawlessly.

However, the real invitation is meant to be shared with others. Supporting users to upload photos, and allowing recipients to directly open the link and load the uploaded photos, cannot be achieved with simple front-end code alone, and is beyond the technical capabilities of a humanities major.

If it's just for practicality, there's an even simpler way: hardcode it! Hardcode it! Hardcode everything into the code! Instead of creating a complex invitation generator and then using it to generate an electronic invitation, it's better to directly tell Professor G the information and let him create a shareable invitation webpage. Therefore, the prompt text was modified as follows:

After several tweaks, a shareable, customized invitation was born. Teacher G seems to have been recently brainwashed by the song "Passing Through the World," trying several times to include the lyrics in the invitation. It only stopped after I reminded it a second time that "Passing Through the World" isn't a song by Yoga Lin. Now that things have come to this, could "The Unending Voice: A Chronicle of Chinese Culture" perhaps get Yoga Lin to cover "Passing Through the World," turning Teacher G's "bug" into a feature? (No advertising here, 100% personal preference.)

In short, after about 10 minutes of tweaking, I successfully obtained a customized wedding invitation that I could share directly with my guests.

Test 5: Generating a 24-point card dealing and scoring system

Beyond tools and educational scenarios, the Gemini 3 Pro can already generate practical game tools, such as a customized card dealing and scoring system for offline gatherings. The prompt consisted of 102 words, and the total time from thinking to completing the code was 124 seconds.

After testing, the main workflow works flawlessly and can be used directly in offline gatherings. Similarly, many offline mini-games that lack suitable tools can utilize the Gemini 3 Pro to customize a set of dedicated gameplay tools.

After testing the Gemini 3 Pro for two days, as a liberal arts student with zero coding skills, besides marveling at "G-Teacher's power," I have some small insights worth sharing.

1. The more abstract the requirement, the more complete the delivery.

Compared to other Vibe Coding products, Gemini 3 Pro's ability to understand "requirements" is exceptionally strong. These aren't the kind of requirements delivered by product managers, which need to clearly define various functional boundaries and exception handling in a documentation document. Instead, it focuses on the core: what I want to do, and what I want to achieve. It can fill in the rest—interaction logic, implementation details, and even design style—for you.

In real life, interaction designers and developers hate it when product managers give "one-sentence requirements" or "I have an idea," but this AI assistant excels at executing such abstract requirements. It especially dislikes users "teaching it to do things," and even when it's asked to make adjustments, a more effective approach is to describe which requirements are currently unmet, rather than instructing it to add a specific button on a specific page.

2. The more bugs are fixed, the more bugs will appear.

However, there's no need to mythologize the Gemini 3 Pro, such as making assertions like "the front-end no longer exists." The Gemini 3 Pro still faces certain difficulties in understanding and executing precise commands, typically resulting in more and more bugs being added as more are fixed.

If you give even the slightest debug or tweak command, it might just crash. For example, in the Pythagorean theorem proof demonstration, the first version of the graph shifted position after being moved. After I requested adjustments such as "the triangles in the square on the second page shouldn't overlap, and the graph boundaries should be more defined," it wrote another 110 seconds of code, delivering the result shown below:

Moreover, once it crashes, no matter how many rollback commands are issued, it won't come back to life; the newly added bugs will stubbornly persist on its corpse. Debugging is something that still requires manual adjustments by users who know how to code. As for users like me with a liberal arts background, the only solution is to start from scratch and have it write a new script.

3. Is everyone a front-end programmer? Or is everyone a product manager?

After the Gemini 3 Pro was released, in addition to setting up my own web pages, I also asked a current front-end programmer and a product manager about their user experience.

A current front-end programmer asked Teacher G to make a card game. After testing the main gameplay, he sent a sweating penguin emoji, along with the hints and screenshots of the game page, to the group chat, commenting, "Okay, I'm out of a job." But from a more serious perspective, he said, "The hardest part of development is understanding what the product and interaction designers are saying and being able to refute them."

The product manager spent two minutes creating a Gomoku (Five in a Row) game webpage for himself, battled the robot for 20 minutes, and ultimately lost.

An hour later, she came to report the good news, saying that she had finally won.

For ordinary users, the most welcome feeling is that tool-type products may gradually move towards the era of "personal applications". This time, everyone can really be a product manager, even if the product only serves one person.

In the past, utility products were often limited by monetization. Only when the user base was large enough or the demand was strong enough could a product survive on advertising revenue or user payments.

Conversely, some needs with a small audience and limited scale are often difficult to fully meet, or they may be met while simultaneously subjecting the audience to persistent, annoying advertisements. For example, if you search for "online Gomoku" in your browser and click on the first link, the advertisement page is larger than the Gomoku board itself.

Searching for "date countdown" and clicking on the top-ranked online webpages reveals an interface with no design whatsoever, while the advertising stickers at the bottom are particularly eye-catching.

When users have the ability to build simple web pages for themselves with the help of AI tools, the last mile of tool products can be completed by the users themselves.

It is no longer synonymous with ugly pages, ubiquitous advertisements, and easily accidentally clicked links, but rather a production process equivalent to creative drawing, content production like producing a line of text, a photo, or a video, a truly consumer-centric capability rooted in the consumer.

Therefore, regardless of past coding experience, you can start boldly experiencing things now. No matter how many reviews you've read, only when you actually write your first prompt and open your first webpage can you clearly perceive the kind of era you live in.