The best bit of ai so far (other than generating testbox bddd testes which then generate code (that is 'ok')... is asking about implementation. We are just rewriting our cfml framework and gettting AI to check best practices and describe what we are doing and asking better approached to do it is invaluable. The other cool thing is after describing the framework and how it works in a workspace for our framework we can just ask AI to generate new modules/extensions, and it builds it out, testbox bdd test suite and all (we've also given links to docs, sourcecode in code generation instructions for ai) That way it fully understands the context of different items, and even understands the dependancy injection, and other custom attributes we use in tags and properties to decorate factories. (edited) There is clearly not enough public good cfscript out there at the moment because most of these engines need a whole stack of instructions in settings just to get good code. So just put that you use fw/1 in your copilot or oather ai code generation instructions and give testbox or other instructions for test generation instructions and you're away. We also have a github gist now with AI instructions and just say 'instructions are at {gisturl} I just created the #C08QZDKHN9G channel so we can continue there

@dawesi I’ve had several blocks from copilot in vscode about not being able to get a response due to public code

even with facile queries that shouldn’t involve any complicated solutions

many of my current projects are fw/1 based both old and new so I’m a bit stuck asking it for help

(this is using claude of any version)

It would be great to have a gist for each framework that can be included in a workspace. (hopefully framework writers can come up with one each) I would as a starting point create a vscode workspace for any work done with a particular framework, then in the instructions settings, point to a gist url with a paragraph of text describing where to find framework docs, source code, and secondly setup instructions in settings for testing with another gist and links to where the ai can get docs and examples from and if necessary also explain testing style. I have a textbox blurb with source url, docs and a couple of good blog post explaining how to use it for bdd for example.

something like:

This workspace uses cfml in cfscript, using the lucee engine version 6 ( docs at : <https://rorylaitila.gitbooks.io/lucee/content/> and <https://docs.lucee.org/guides/cookbooks.html> and <https://docs.lucee.org/reference/functions.html> ), we use the fw/1 cfml framework ( source at : <https://github.com/framework-one/fw1> and docs at <https://framework-one.github.io/documentation/> ) using AOP ( <https://framework-one.github.io/documentation/4.2/using-aop-one/> ) ...

then go on to describe any major options in your setup (like AOP/DI/etc) so that it can get the nuances of how you do things. I also like to put a directory structure description and/or point to a doc page that has that also.

I've had no blocks, I don't ask for public code, with AI you need to ask questions about code generation, not code copying... I ask questions like:

build me a component to send events to clickhouse

or

use rereplace to convert a version from `v12-0` to 12.0

or

in this workspace, look at all modules in the /modules folder, and ensure they all have "tc:schema" parameter in the component attributes, of not set a default of version 4.3, and report back to me a list of changed files.

or

analyse my code and find any code that is inconsistent with our coding standards at : {url}

You can now also store your instuctions in

copilot-instructions.md

files for copilot. Customize chat responses in VS Code

I'd love to see someone sync all of the frameworks to a single github account and turn on the

Allow GitHub to use my data for AI model training

and

Allow GitHub to use my data for product improvements

checkbox under 'privacy' so that their repo can be added to training models for cfml. Unfortunately you can only turn this on at 'account' level from what i can see, wish I could find a per-repo setting on this.

I'm finding AI useful (some areas more than others!) but you can not trust it, everything needs to be reviewed. So it's a bit like a very keen computer science graduate has just joined who knows a heck of a lot of theory but in hasn't got any real world experience of working with code that has evolved over years if not decades! Good article from Stack Overflow which isn't too far off from my own thoughts https://alonso.network/the-recurring-cycle-of-developer-replacement-hype/?lid=ufkcltqfo3xc

I posted this on lucee.dev but will repost it here: Copilot overage is FREE with NO RATE LIMITS (see below how) So now you can hit rate limits, but ONLY if you do not set a balance for overage, set a $5 balance (and have a payment method) and you can then CONTINUE FREE to use it BEYOND your quoto. I have zero billing...

Hit the limit 10 days in then added more budget, then realised this 'hack' to get FREE UNLIMITED copilot (during pre-release) Make the most of it while you can, you also have the choice NOT to contribute your code to the learning base if you need to have it 'internal'.

@1andypeterson has left the channel

Been doing a lot of Cursor demos for engineers at work. One useful tip: try setting up your @Docs in Cursor pointed to appropriate documentation. I wonders if someone has published an llms-full.txt for CFML yet…

I have a .github/instructions/cfml.instructions.md in vscode which co-pilot reads to give it context

if you are using copilot, it suggest best format is markdown. I get copilot to write it so it can best read it. I might try the @docs or /docs annotations also to see how that goes

Yeah - I've had success by writing an markdown file and then asking copilot to read it and ask for clarifications so the markdown file gets updated my both AI (asking questions) and me (providing answers)

@Matt Jones has left the channel

has renamed the channel from "ai-code-generation" to "ai-assisted-coding"

set the channel topic: Discuss the use of AI to develop CFML programs: share your tips, tricks, and workflows!

set the channel description: A place to discuss how to develop with CFML using AI: LLMs, editor integrations, CLIs, prompts, context engineering. What has worked for you and what has been less successful?

Following on from the discussion in #C06T99N9G : I use (GitHub) Copilot Chat in VS Code for everything. It switches between models automatically, depending on what it needs to do, using cheap models for basic tasks, so you only burn token on hard tasks. I haven't exceeded the $19/month subscription yet, but I'm not doing hardcore development every day these days. I'll be honest, I haven't had it do anything with CFML tho' -- just JS, Clojure, infrastructure/configuration, and documentation 🙂

If you're wondering why you've been "invited" (added) to this channel, it's because you've previously mentioned in #C06T99N9G that you use AI to help develop software, so we're going to try to encourage more sharing of workflows, skills, etc here in a dedicated-to-AI channel...

Thats @seancorfield

Is anyone using a cfml skill? I have had excellent progress with Claude code, 1000x faster then me coding. I omly have 1 active site I maintain in cfml, but I do a bunch of tools on the side.

Anyone tried Kimi k2.7 code ?

Curious as to people's thoughts on building with CFML with AI when it's so easy now to build with alternative (and perhaps more AI-friendly?) languages? Has anyone started branching out now that the learning curve is vastly reduced? For me, I'm working on a monolith legacy CF app with very little tests, so rebuliding it from scratch in another language is not reasonable even with AI, but I am working through a modernisation effort screen by screen, building tests as I go, and AI has helped immensely with that. It occurred to me that when I'm in a better place automated-testing wise, it may not be a stretch to start building in Go/Java/Rust and avoid all the little runtime pains that CF can produce. In saying that, using AI I've found that those little mistakes tend to be vastly reduced or even non-existent (especially in the last 6 months or so) - perhaps the compile safety that those other languages offer isn't as important when the AI's CFML code quality is already pretty good?

It's probably just me. I find all the new, and ever changing naming conventions for AI to be annoying. Things like open-weight, inference, multimodal, agent, model, frontier, and harness. All annoying. Maybe I need a Snickers bar.