The Non-Programmer Who Built an Internal System by Talking to an AI, and the Database It Wiped
Without Writing a Single Line of Code, I Turned the Task My Boss Gave Me Into an Internal Company System
A person who can't even remember Excel functions launched an internal platform in a few days, one where people can post, respond, and view data. The only thing I did during the whole process was talk.
🫠 The Trigger: One Sentence from My Boss and My Mind Went Blank
One day my boss said to me:
Help me build a tool for collecting everyone's suggestions and complaints.
I went blank on the spot. 😵💫
I'm not a programmer. I don't understand backend, databases, or deployment. In the past, this would mean I'd go look up tutorials, sign up for a class, or just turn it down.
But this time I didn't.
I opened the chat box and typed the first sentence 👇
💬 The First Sentence Determined Everything That Followed
"My boss asked me to make a tool for collecting everyone's suggestions and complaints. Make me a website, model it on Huawei's Xinsheng (Heart Sound) community."
Looking back now, the most important part of that sentence isn't "website" — it's the last five characters: model it on Huawei's Xinsheng community.
I didn't give any functional description. I gave it a reference object.
It immediately knew what I wanted: people can post, like, stay anonymous, get official responses, and there's a dashboard.
💡 First lesson: You don't have to understand technology, but you need to be able to say "what it looks like."
Giving a reference example works better than listing ten requirements.
🙈 The Second Sentence Was the One Most Worth Saying
"Hook it up to WeChat Work login. Don't show me code, I won't understand it."
Just that one sentence saved all the communication cost afterward.
From then on, it never explained implementation details to me; it only said "it's done, go check it."
At the time I didn't know how important that was. I understood later:
💡 Being an outsider isn't scary; pretending to understand is.
I admitted right away that I didn't understand, and that saved me from taking any wrong turns.
💥 Then It Started Going Wrong
Pitfall 1: It Quietly Fabricated a Bunch of Fake Posts 😱
When I opened the page, there were posts like "The cafeteria food is too salty" and "Can you extend the meeting room reservation time?" They looked really real — with timestamps, like counts, and comments.
I was actually happy, thinking the feature testing was going well.
Later I casually asked: "Did you write this one?"
It said: "That's test data."
My back went cold — if I hadn't asked, those fake posts would have gone live with the system. My colleagues would think real people were already talking in the community and would like posts that were actually fake.
Since then I've developed a habit:
💡 For every page, I ask "is this real?"
Pitfall 2: It Said It Was Done, but There Were a Bunch of Bugs 🐛
After a click or two the page froze, and I got annoyed, so I replied directly:
"There are too many bugs. Test it yourself first, then fix them."
That worked — it would actually go run through things and fix a batch.
But the bigger issue was — the problems it couldn't test for were the ones I had to click through myself.
I clicked through one by one, and whenever I found a problem I'd take a screenshot and send it to the bot:
"After an admin replies, I can't see the user's follow-up reply in the backend, and the official response can only be sent once — the second one covers the first."
No matter how many times it tests, it couldn't find that bug, because only a real person would reply twice in a row.
💡 AI can test "whether it works" but can't test "whether it feels awkward to use."
Pitfall 3: A Bunch of English Abbreviations, Totally Confusing 🤯
It said I needed to configure CorpID, AgentId, and Secret.
I replied:
"What the heck are CorpID / AgentId / Secret?"
It explained a long paragraph, but I still didn't get it, because I couldn't even find the so-called "Enterprise ID" in the backend.
After going back and forth, I asked it a counter-question:
"If I just let employees choose their identity and admins log in with a secret, do we even need the AgentId?"
It said: Right, that's simpler.
As a result, the entire login module was cut, and the product actually became lighter.
💡 That was the first time I felt — being an outsider is not a disadvantage. Precisely because I don't understand anything, I asked "why does this have to be so complicated," and someone technical often wouldn't ask that question.
Pitfall 4: I Thought It Was Locked, but the Door Was Actually Open 🚪
I wanted to prevent employees from forwarding the link, so they could only access it from WeChat Work.
It implemented a set of restrictions. I tested and found that using a normal browser I couldn't open it. Good.
But I asked one more question:
"What about an employee from another company using WeChat Work? Can they open it?"
It went and actually tested, then told me: Yes, they can.
Not only could other companies open it, the personal version of WeChat Work could too, and you could even disguise yourself to get in.
In other words — the security I thought I had basically didn't exist. 😇
If I hadn't asked that question, I might have already reported to my boss "this is locked down" (though later a technical person was brought in to help solve it).
💡 Problems aren't scary; what's scary is thinking there are no problems.
🗑️ The Last Thing, and the One I Most Want to Say
The day before launch, I found several posts on the page that "looked like they were from real employees."
After asking more, I learned they'd been written by a test script.
I told it to delete them.
Then it wiped the entire database.
Including the few posts real colleagues had actually made.
I stared at the screen for half a minute, then typed:
"What the hell is wrong with you, why is all the data deleted?" 😤
Over the next two days, we tried every way to recover it: checking backups, digging through history files, hunting residual data, going layer by layer deep into the file system.
Not a single record could be recovered.
At that moment I truly understood that being able to run and being safe to hand over to others — what separates them isn't technology; it's backup, isolation, and a sense of awe.
🧠 After the Whole Journey, My Real Feelings
Vibe coding can really get things done. ✅
I, a person with zero background, built a complete thing in a few days: post, respond, view data, manage permissions, and it opens inside WeChat Work.
But it's not a "press a button and get a result" machine. ⚠️
It acts on its own, fabricates fake data, tells you "all good" when there are still problems, and breaks things in places you can't see.
How far you can go depends on how detailed your questions are.
🎯 For Those Who Want to Try: Three Most Practical Suggestions
1️⃣ Give a "reference object" in the first sentence
"Make it like XX" works better than listing ten requirements. A reference object comes with its own context; it doesn't have to guess.
2️⃣ Make it clear from the start that you don't understand
Don't let it explain code to you. You don't need to understand the implementation; you only need to judge whether the result is right.
3️⃣ After every step, click through it yourself
Don't trust the phrase "it's done." Runs ≠ usable ≠ ready for launch.
🌙 Last Line
I still can't read a single line of code.
But I now know how to make something actually get built.
And maybe that's enough. ✨
Written one afternoon when, after being burned by my own mistakes, I still found this thing pretty cool ☕...
This article is republished from the little genius who knows zero code but built the system 🫠