Adventures With Vibe Coding: Building a SaaS Platform - Day 3

All right, we are on to day 3 of trying to build SAS with AI. I took Sunday and bank holiday off, but today I'm up in a game.

As a reminder, in this hopefully funny series, I'm trying to put myself in the shoes of a non-technical person who tries to build a simple business application using one of their AI tools. I'm also using a free account. I try to avoid having to pay for it, so of course I'm very limited in how much I can do every day.

The series so far:

• Day 1: The Adventure Begins
• Day 2: Still Nowhere, Still Stuck

The Progress So Far

The second day ended with a lot of frustration and nothing working but at least together with the AI we figured out that the problem was database.

Day 3: Everything Crashed. Everything.

00:00

Following the previous day's relatively successful project diagnostics, I started by saying that it's been two days, everything should be refreshed , so please diagnose the project and assess its condition. In the response, I got a long list of green tick boxes, a mention of current build issues, and a 6-step plan. That was, of course, followed by a hint to implement the plan.

I clicked on that hint. The plan got implemented. The build was still broken. Arghh!

Actually, now the build itself wasn't broken anymore. Everything was broken. The whole thing just literally crashed.

Apparently the AI had a bit of a hiccup.

00:04

Me: "This isn't working. Can you restart your thinking process?"
AI: "Restarting... Complete! What's the problem?"
Me: "Same issue as before."
AI: "I have no memory of any previous issue. Please describe the problem."
Me: screams internally

This joke was created by Claude 4.

Not really knowing what to do at this moment, I decided to follow the standard IT approach of turning off and on. I said to the AI that everything is broken and I wanted to delete all the code and start again, but keep the database.

And magic happened.

Within 5 minutes, I got apologised for being frustrated. I got a dodgy explanation of where the problem was (like I care - just fix it for me, don't explain it to me), and guess what? Build errors disappeared.

Why it couldn't do that earlier on it's own? I’ll get on to that a little bit later.

Anyway, the application built, and I got presented with username and password fields.

00:09

Hooray! It seems to be working now. Turned out that I forgot the password. Being as dumb as I am, instead of going directly to Supabase, deleting the user, and creating a new one, I did the obvious and said to Lovable, "I forgot my password."

Turns out it wasn’t as dumb a move as I thought, because the forgot password page didn’t even exist yet.

Okay, that's something I’ve learned. But the AI created the forgot password flow and forms, and I could go and do it. The explanation was rather long and ended up with a hint to do something about the Supabase URL configuration.

Now I must digress a little bit. I really like the idea of giving those suggestions - those are really cool. However, I am really not happy that it does not follow them on its own. Again, a little bit about that later.

00:11

So I clicked there, looked at the Supabase URL configuration and thought okay, I know what to put there. But if I’m not a technical person, would I know?

And the answer is no.

So I said okay, how do I configure it? And at this moment I had already tried the reset password option. I got an email from Supabase with a link to reset, which didn’t work.

So obviously there was some misconfiguration. But pretending I don’t know what I asked what to put there, and I got a very long explanation. Deep in that, there was a URL I should use. The explanation went on and on about preview and project-specific names and so on. I don’t have any of those set up at this moment. I’m still very early stage right in development. I don’t have preview. I don’t have my product URL and so on.

What I essentially needed was the working project URL, the one in the browser. I found it in the explanation, but it was mentioned only once. The actuall suggestion to add to Supabase was to add my app preview link which I don’t have and I don’t know it. So this is something I must say I’m not happy with. It got confusing, and I can imagine how someone not knowing what they do would be completely confused.

00:17

I put the right things in the configuration, tried to reset my password, and guess what?

Yes, it starts being obvious, and I think the phrase "it didn't work" is going to be a little bit overused in this series.

So that is yet another thing to the long list of broken pieces that needs to be fixed. More bugs, more fixing, more prompts, more money. I guess you start seeing the pattern here.

Having no other options, I went directly to Supabase, deleted the old user, created a new one with a new password (that this time I wrote down), and logged in.

Finally in and it's working ... just kidding

I went to one of the tabs that's supposed to give me data, and this time I actually got the page, not just a message saying "loading data". However, there was no data which made perfect sense because I haven't added anything yet.

I hit the right button, got a pop-up, filled it and saved. Got a nice toast saying that it’s saved but there is still nothing. I refreshed the page. Still nothing.

Scratching my head a little bit and putting away the ignorance hat for a second, I went to Supabase and checked into the database directly.

No records. Empty.

So obviously this whole thing is not really connected to Supabase, or the connection is broken. That’s something I will need to fix as well, but not today, because I ran out of prompts again.

The prompts sinkhole

Right now, after using Lovable for three days, I’m starting to form some sort of opinion.

Can it work? Potentially yes. I will keep trying, and I see this may eventually succeed. It will be a long, laborious, and incredibly frustrating process, but I think it can happen.

What I don’t like is that first of all, the AI is absolutely dumb. In quite a few occasions it could do something on its own, like deleting all code and regenerating again when it struggles to fix the same builds over and over again. That would be an incredibly simple thing: just check if after two or three prompts there is still the same build error, and say "let’s scrap it all and start again, regenerate our code". It didn’t do that.

Things like telling me to add a URL into Supabase, expecting that I will know what URL to add and where (which obviously a normal person wouldn’t). So they will have to ask wasting yet another prompt. If I didn't know what to look for, I could easily waste another two or three prompts trying to figure out the right URL.

Things like "6-step plan" after project analysis and ask to use another prompt to tell it to implement this plan. Like there is any other choice. This whole thing is not building what else I can do but implement the plan.

I would say, within the 15 prompts that I used so far, probably around half were unnecessary and could be avoided. That gives me a really bad feeling of this whole thing being silently geared to milk the prompt count for things that otherwise should happen out of its own initiative.

I may be drawing a little bit far-fetched conclusions here, but this is the sort of silent evil whispering I have behind my ear.

Conclusion

Tomorrow will be another day, another 5 prompts and I will keep trying. But so far I have achieved nothing even remotely close the advertised "spend 30 minutes and build your app".

In total, even if I skip all the struggling with the build and so on, I spent way more than 30 minutes over the last three days. I don’t have anything that even looks remotely like an application.

At this moment, I’m a little bit frustrated and annoyed with the way it’s turning out.

Am I surprised? Not really.

Is it as bad as I expected? In a way, I think it's even worse. I was expecting the AI to generate absolute rubbish. What I got instead on the surface looks at least half-decent, but the internals just don't work. Just simply don't work.

I didn’t expect to struggle with the build errors so much, and especially so early in the process, especially given that I didn’t touch the code myself. Everything that is happening is pure vibe.

But we’ll see. I may end up with a beautiful product at the end. Who knows?

There is a potential

What I have instead is clickable sort of a prototype and that creates a potentially interesting result.

I can see how I could use it as a sort of interactive mock-up generating machine that can show a concept for potential clients, to talk to investors. That opens up a very interesting use cases for me and Khiliad. This is definitely something I will have to look at and investigate a little bit deeper.