New to Lucee, not a developer, trying to improve an existing app with AI — looking for practical advice

dev
1

Hi everyone,

I’m new here, and pretty new to Lucee as well.

I’m not a developer, and I don’t have a technical background beyond being reasonably comfortable with digital tools. A few years ago, for a startup project, I had a web app built by a small local software house. The app was built in Lucee/CFML, which is also the main reason I’m posting here. It was one of those small teams that move fast and get things done, but not always in the cleanest or most structured way. For a mix of reasons — partly budget, partly personal circumstances — the project ended up stalling and never really went beyond a decent MVP.

A few weeks ago, mostly out of curiosity, I decided to see whether I could get the app deployed myself with the help of ChatGPT, since the original developer had taken it offline. Somehow, I actually managed to do it. After that, still more as an experiment than anything else, I installed VS Code and the Codex extension and started making a few small changes, mostly on the UI/UX side.

Step by step, I’ve gone a bit further than I expected. I’ve started adding features, adjusting existing logic, and even touching some architectural parts of the app.

So far, things seem to be working: I’m making changes, testing them, and keeping everything versioned on GitHub. What I’d really like to ask is this: given that I’m not a developer, do you have any practical advice on how to use AI coding agents in a smart way, so they don’t gradually produce poor or messy code? And are there any critical areas I should be especially careful about?

I’m also aware that asking a developer community for advice on using AI for coding may not be the best opening in the world, so I hope this doesn’t come across the wrong way. I’m not trying to replace proper development expertise — I’m mostly trying to understand how to avoid making a mess while learning and experimenting with an application that already exists.

The app itself is a small rental management system that lets users register and manage inventory, availability calendars, and orders.

Thanks in advance to anyone willing to share a few thoughts.

1 Like
2

Just a quick thought on this - interesting use case! I’ve been using Claude Code a lot lately (I’ve been doing CF for 20 years). Mostly I’m working on an app which is an industry-specific media solution. Claude has helped me a lot to make faster progress, refactoring old code and implementing new features. 80-90% of the work is amazing; the remaining part is sometimes tricky. So here’s what I’ve noticed:

  • Shit in, shit out: if your codebase has no consistent patterns, coding style, or security-minded coding, the AI will probably continue to mirror that — parts will be fixed, others will remain bad.
  • Debugging can sometimes be hard. If you’re not able to describe the problem correctly, you’ll end up in an endless loop where the AI tries to fix it.
  • Try to test as much as possible before deploying live.

Maybe you could get somebody experienced to review the code and changes and give you some advice from time to time?

1 Like
3

some good advice, things I’d suggest…

  • code should tell a story, you’re already getting somewhere, monitor the code changes using github desktop or similar
  • if you don’t understand, or it confusing, challenge the AI to explain it
  • if you aren’t already, use tests, that’s the best workflow with AI
  • use spec driven development, before making any code changes, ask to write up a md file about the problem, then to investigate, present options
  • LLMs love to write overly defensive code (too much catch try, elvis etc) let code fail, otherwise, it runs around trying to solve problems it’s hidden (my moto fail early, fail hard)
  • good documentation and logging really helps both the AI and your future self
  • keep the focus tight, work on one thing at a time, write out a spec for the next thing you want tot address, when there’s something you discover working on something else

a few of my favourite prompts?

  • “grumpy review”
  • “pros / cons / concerns”

Learn from it’s mistakes, create an AGENTS.md / CLAUDE.md file which you can prevent repeating mistakes, by giving it rules

5 Likes
4

“I installed VS Code and the Codex extension and started making a few small changes, mostly on the UI/UX side.”

“I’ve started adding features, adjusting existing logic, and even touching some architectural parts of the app.”

“I’m making changes, testing them, and keeping everything versioned on GitHub.”

Hey guess what … you are a developer! :smiley:

We all start somewhere and it takes a lifetime to learn that there’s always more to learn and nobody knows everything no matter how long they’ve been a “developer”.

I’m not even sure what the definition of that is anymore, but “I’m not trying to replace proper development expertise” shows that you’re already on the right track!

AI has humbled me by making me realize I wasn’t using all the best practices. I think it’s more important using it as a learning tool than a coding tool. Windsurf’s Supercomplete feature is great because it’s more like pair programming than just having AI write all of the code. Also I like to make AI review not just my own code but the code that the AI just wrote for me! And having as complete a test suite as possible helps to prevent breaking changes.

2 Likes