Lucee in the age of AI

dev
1

I’ve been exploring - Any thoughts from the devs here?

One of the biggest unsung advantages of the ColdFusion/Lucee architecture in the AI era:

:mag: Its page-based structure — with self-contained .cfm and .cfc files — makes ColdFusion AI-friendly by design, even if that wasn’t the intention.

:bulb: Why Page-Based CFML Is Perfect for AI Refactoring:

:white_check_mark: 1. Logical, Discoverable Units

  • Each .cfm file usually does one clear thing: display a form, run a query, send a response.
  • You can give an LLM just that file, and it has all the context it needs — no hunting through 20 folders or classes.

:white_check_mark: 2. Minimal Dependency Chain

  • Unlike frameworks like Django, Spring, or React, there’s very little architectural scaffolding.
  • Less boilerplate = easier for AI to understand and manipulate directly.

:white_check_mark: 3. Simplicity Over Purity

  • ColdFusion has always prioritised getting stuff done over theoretical elegance.
  • That’s precisely the kind of pragmatic, explicit logic that LLMs thrive on.

:white_check_mark: 4. HTML and Logic in One Place

  • With CFML, especially in legacy apps, your UI, control logic, and business rules might all live in the same file.
  • This violates MVC purism — but gives AI a richer local context to reason from.

:wrench: Workflow That Works Beautifully Now:

  1. You need to refactor an order processing script.
  2. You grab the processOrder.cfm file (or chunk of it).
  3. Paste it into ChatGPT.
  4. Ask: “Make this into a reusable function that returns a struct with success, message, and result.”
  5. Boom — the AI has everything it needs, right there.

No:

  • ORM mapping
  • Container bootstrapping
  • Class scanning
  • Dependency graphs

Just code. Just logic. Just output. And not many/any functions required…

1 Like
2

I’ve written some posts in the last month here about this.

The secret sauce in AI dev is the INSTRUCTIONS file which the ai (hopefully) uses to craft your code. I’ve got a directory of instruction files and in my copilot-instructions.md I just point to those docs using a markup document (markup is best to use for ai apparently)

One for CFML coding standards (based on the ortus one)
One for my TeamCFML cfml framework (our 25 year old framework)
One for Testbox (for testing)
One for my database framework (i call diimes)
One for my sencha ExtJS layouts (HelpfulUI)
One for Siesta (js/ui testing framework)
One for my screenshot testing app that takes screenshots of my apps during unit testing (then hash the image) and compare image to previous version to prove interface is the same (or different with new feature in that area).

Then just tell ai what to do, and add new items along the way, or change the rules in your doc.

and I use a coldfusion port of docsify (that AI wrote in 5 seconds) to do all my docs in markdown as docsify collections, then got AI to write a documentation portal that nests all the docs from above. Docsify is super easy to use with ai also.

Copilot is amazing when you understand how each LLM works.

Check out:

Heads up : overage for github copilot is currently not charged - Tools and Libraries - Lucee Dev

3

So exactly the opposite of my post. My feeling is that the only point of using many of these legacy frameworks is to match the boilerplate of the past.

What is the point of this in an AI Native world?

4

boilerplate has it’s place… look at 99% of apps with the same look and feel.

but it does seem like backend is more for creating headless systems now, and the divide to ui is data.