You’re absolutely right: Lessons learned vibe coding our asset management system

For the last little while I’ve been playing around with Replit to build a few apps for my coffee roasting business East Coast Roast. In this article I’ll break down some lessons learned over the last few months. You hear a hell of a lot of opinions about ‘vibe coding’, ranging from ‘it’s impossible to build anything of use’ to ‘it’s going to replace all developers’. This article is all direct experience so hopefully it helps cut through the noise.

First up, check out one of the apps I’ve built, an asset management app to help us manage our equipment. Here’s a quick run through of the app and below I’ve listed the main features.

Here are some of the features:

• Dashboard with asset summary including $ value, assets requiring depreciation, assets with upcoming service required, a timeline of recently modified assets and more.
• Assets area displayed as cards, or rows.
• Ability to filter assets based on a range of fields including service due, depreciation method, location etc and any other custom field needed.
• Export all assets to excel file.
• One click print a summary sheet for each asset, to use to stick to the asset in the warehouse.
• Locations area to manage locations where you might keep assets (your own locations, or customers / external storage etc).
• a Maintenance section enabling you to log maintenance of assets and create automated rules on what assets are in and out of service.
• Balance sheet section enabling you to group assets into accounts for the balance sheet so you know what the value of each grouping is (for your accounting system).
• Full depreciation system that lets you create depreciation rules for assets and apply them so you can appreciate assets in various ways.

All built with AI, not looking at a single line of code, I don’t even know what programming language this is written in. This is live, and we are using it now.

Here are some lessons in no particular order.

Cope

A lot of developers are trying to fathom what the world looks like when a very difficult skill that they’ve spent their life cultivating, is no longer needed. I’ve seen so much negativity around vibe coding online that I’m kind of surprised I was able to build something useful with it. But I have built something that lots of people are paying decent monthly fees for, and I’ve done it without a developer.

I think you need to try something yourself rather than pay too much attention to what people are saying online. There’s an obvious conflict with people who do this for a living, who are trying to cope with the fact that they are being replaced. I built this on Replit, have a play around with it and make up your own mind.

Explosive improvement

I started playing around with Replit a few months ago. I tried to build a mobile app that tracked site visits for sales reps. I built the back-end in Replit. Replit has 2 modes, an Assistant that you can ask questions and it can generate code for you that you apply, or an Agent that just goes and does it. I built the sales tracker app by asking the Assistant questions and then generating the code slowly either via Replit or external AIs like ChatGPT or Grok. Whenever I used the Agent I’d freak out because it would get it wrong constantly.

The mobile side I built in Expo using VS Code and Cursor on my mac. Replit struggled a lot as did VS Code and Cursor. It was awesome to build something without knowing how to code, but it would break constantly and send me in loops.

When Grok 3 came out, I started pasting code into Grok and it would get it right more often than not, but the process was extremely slow.

That project was complex because it was a mobile app, not just a simple web app so I shelved it and started to work on 2 web only apps using Replit solely. A few months passed and the improvement has been so big, that to build the new asset management app I didn’t use the Assistant at all. I used the Agent for everything and it just built the entire thing. I didn’t even have Github set up, no backups, constant overwriting of code, the agent did it all. I didn’t even look at the code. I didn’t even look at the files area that showed the code. I still don’t even know what language this app is written in.

But the beauty of Replit is this is all hand coded stuff, so I can take all the files away if I like, it’s not like the old school drag and drop coding system where you have no idea what’s going on behind the front end. I could take this app and host it elsewhere, or get a developer to work on it with me. This is a proper web app, not a toy.

All of that to say that the improvement, just in a few months and a few new releases of the various AI models, has been explosive. We’ve essentially gone from unusable, to replacing multiple people’s jobs in a few short months. They’ve spoken about the ‘no code’ revolution for so many years. We are finally here, and who knows how much improvement there is to come if those changes have happened that quickly.

No skill

A lot of what I’ve seen online around vibe coding is either “you have to be a developer to use it”, or “you need to be skilled in certain prompts”. There was talk of some kind of vibe coding profession emerging.

I’m not convinced. I built this whole app drinking wine and angry-typing yelling at an AI. I’m not sure I’d call that a skill. I didn’t use a designer, or a developer, or a project manager or anything. I just mashed buttons.

Things it specifically struggles with

There are some things to me that I noticed it struggled with.

1. Timezones – I’m not sure if all developers struggle with this, but the AI really struggled with figuring out the difference between the server time and the times to show in the app. This one didn’t impact the asset system but I’ve also been building another app (more on that later), and I ran into lots of brick walls with timezones.
2. File duplication – Quite often the AI would be convinced that it fixed an issue and couldn’t understand what I was seeing, and the whole time it was working on the wrong file after creating a backup for a file and not realising. Knowing this should help make things a bit more efficient in the future.
3. Security – I noticed there was very little attention paid to security. It created logins without hashed passwords, created systems for automatically entering passwords and lots of other things. At some point if it’s going to be a system that’s available online, some attention would have to be paid to security. I reckon it would be up to it, but you’d have to make sure you consider this when you build it, you cant just let it do it’s own thing.

Also worth noting, as mentioned before if you are building a mobile app things get a LOT more complicated. This is a web only app and that simplifies things a lot.

Can be frustrating

It can certainly be frustrating to build something with AI. Here’s one of our conversations.

While there are moments of frustration like this screenshot when you build with AI, it’s often a lot more frustrating to build something with human developers.

I’ve been working with developers for over 20 years, and for the most part, the process is torture.

It’s fun

Overall I’ve loved building this app with Replit. I’m kind of addicted to it. It’s so awesome to be able to build features so quickly and do it all yourself. I was at work the other day, after climbing up a ladder to our mezzanine to look at a coffee machine, only to realise that I knew nothing about it. I had to take off the water waste grate to see the serial number and go back to the system to put it in, to learn the history of the machine. I thought, ‘we really should have a piece of paper that lists the asset info and have it stuck to the machine’.

I went back to my desk, opened Replit and asked it to build the feature. It took a few minutes and the app now had a button I could click to generate a PDF that listed the asset details, notes and full history. Now we stick these notes to our machines.

This is so damn satisfying.

I’ve always felt a deep dissatisfaction working with developers in building web apps. It’s just not possible for the developer to work as fast as my brain works in thinking up ideas. But now it can, and it’s liberating and awesome.

Costs

Replit only cost $25 US to use which is obviously an insane bargain. But they do hit you with usage charges if you use the agent a lot, it can add up quickly. I got some bills for this app for a few hundred dollars and I wasn’t expecting them, I kind of thought it was all included.

I emailed Replit support to ask about the extra charges, and had a back and forth email exchange with the support person about what it all meant. It turns out using the agent can rack up $ pretty quickly. There is very good transparency on fees and usage data, as well as features to pause when you hit your pre-set limits. But it’s worth knowing that if you are constantly using the Agent to build, it’s going to cost money.

It can be particularly frustrating if the AI is going around in circles breaking stuff and you know you’re paying for it. Much like the anxiety you feel when you are paying a human developer to build something for you and they charge you an hourly rate to build something that isn’t what you asked for. I do not like this feeling and didn’t like feeling this using Replit.

But finding this out, I realised a few things.

1. The email support from Replit was so good, in particular they were so fast to reply to the emails (i.e. instantly), eventually I figured out that this couldn’t possibly be a human. And of course it wasn’t. I’m pretty sure this is the definition of the touring test.
2. I was super frustrated when the AI went in a few loops to build a feature and I was charged $25. But then I realised I’ve built an app that would cost tens of thousands of dollars to build a few years ago, so the $25 was a bit more palatable.

Risk

Of course I’m just building an internal app for us to use that doesn’t have much risk associated with it. I haven’t looked at the code, it’s possible it’s complete dogshit, I have no idea. If I was building a SAAS startup that people relied on, you’d have to be a bit more careful.

I think one of the hardest things to navigate in this whole thing is that people like me are going to start building this stuff, and actual developers who know the code are going to hate it. But if these apps are going to be used by companies with sensitive information or high risk data, you’re going to need a developer to have some idea what’s going on with the code. How will you do that if developers are hating this technology? How many developers can you trust to give you real feedback on your code when you know they have a huge conflict?

Perhaps developers who embrace AI will emerge and provide a super valuable service? Or we move towards companies building their own apps to use themselves, instead of SAAS companies dominating the whole space. Not sure, but it’ll be fun to watch.

Let me know if you have any questions or thoughts in the comments below.