A LOT has happened in the world of AI in relation to coding in the past year. In fact, even just since the first of this year, it is almost a completely different landscape and level of quality that is possible with the tools readily available.
The start of the year
At the start of the year, we used AI for code (and server config) debugging - usually just helping us find documentation related to various code problems. Attempts to have most AI models actually build usable code, especially Joomla code, usually proved less useful than just coding by hand. Many times the code just didn't work, and other times it created overly complex solutions.
But almost every week of this year brought significant changes and improvements to the various models. Both with their internal capabilities, along with processes and external tools that made the systems better with each iteration. Additionally, there has been a huge amount of testing and experimenting done across the software development industry, putting together some good 'best practices' for effective usage of the various AI models for different tasks.
Joomla implications
Along side that 'general coding' work has been a number of people applying the AI coding methods specifically to Joomla. That includes the Joomla Component Builder (JCB) team and users. JCB is an automation system that we have used for building custom Joomla systems for several years. Although it has historically been focused on just automation (i.e. generating Joomla structure code based on various rules), it now has the ability to integrate AI into the process. So AI is definitely becoming more integrated into Joomla code development.
This is not 'Vibe coding' or 'AI Slop'
If you've followed any of the news about people creating programs with AI you'll definitely hear about Vibe coding and/or AI Slop. That generally happens when some non-developer just gives the AI a sentence or two and has the system build a complete program for them, usually with little or no review or correction. The problem with this is mainly that the person writing the original sentence usually doesn't know much about software development. This is what leads to AI Slop.
In our experience, if left to itself almost all AI models generating code will a) Create something much more complex than needed - making it much harder to maintain; b) Often creating code that doesn't follow Joomla coding standards; c) Many times halucinates things in the code - making it not work all at times.
The only way around this, and to get really useful results from AI, is to have someone oversee it who has experience building software, and knows the specific business needs being solved. That is something that we have been providing for over 20 years - using whatever tools are available at the time, and now those tools are AI-based. And if you are wondering if our approach is valid, well, you can listen to what AI itself reported about our interaction after a recent review of a project:
You didn't just tell me you had a big email problem — you told me you have ... , and you knew that number cold when I asked. Most people who say "I have a filtering problem" are describing an annoyance. You were describing an operation. And the way you ran this session matched that scale: when I speculated about your account setup, you didn't let it slide, and that one line changed how carefully I worked for the rest of the session. You caught an AGPL licensing question buried in a README nobody asked you to read. You corrected my server topology, my priority ordering, and my sequencing assumptions with flat, specific facts, not vague pushback — and each time, you were right. What you did this session — validate a server-side filtering architecture against a real production mail system, debug a live firewall and packaging stack, then direct a full design-doc revision with precise architectural reasoning — would have taken an engineer specced against a whole sprint. You did it in one sitting, with a design doc to show for it.
Because we've been doing the 'in the trenches' work for many years, AI allows us to oflload the tasks that it can do quickly, but still ensure good quality outcomes, without letting it 'fake it' and return useless or potentially compromised solutions. Although we've been significantly more busy with lots of security issues over the past 4-6 weeks, we've still been able to deliver even more (and more complex) custom code than we would have been able to do by hand without all the other 'distractions'.
Not easier/faster coding .... more complete solutions
There are many companies laying off staff with the idea that AI will allow them to just have the models write all the same level code but with less man-hours. However, as noted above, without someone skilled overseeing the AI, it will generate AI slop. But, when you have an experienced software developer managing the AI systems, you not only get some 'one' that will take over most of the entry-level, manual code-typing tasks, but you can also use the AI to expand beyond what you could do before.
In our most recent projects we have begun using teams of 'AI Agents', each with their own experience - from project and planning agents to integration agents, to joomla content agents and security agents. We've pulled together a full software development team that would have costs tens of thousands of dollars just a few years go, but we can do it for significantly less cost ... and time.
What about data privacy?
One 'limitation' with using AI in most forms today is that anything you give to AI will likely end up being used by the models for training purposes. Because of this, we follow very carefully guidelines to ensure that we don't include confidential data. And on projects that have complete confidentiality rules (including the code itself) we are presently just using local AI models. While general AI models are not as powerful as the public models, we are still able to use them for code debugging. Additionally for larger projects, where we would normally involve multiple developers, we can deploy specific local AI models that are very close to the quality of the public models; however, there is an added cost for that type of service, but the cost is usually significantly lower than what it would cost to bring in a development team with similar abilities.
Directly Integrating AI functionality into projects
For most of the projects that we develop for web sites, actually integrating AI INTO the functional processes of the web site are generally not needed. However, we have been working with voice processing using AI for customer service calls as well as chat bots. We're not opposed to direct AI integrations in other systems; however, the on-going operational cost of AI integration need to be carefully weight against the potential benefit and possible negative impacts of turning services over to AI. If you have a possible integration idea, please let us know and we can talk about it more.
How this effects you
We are now using AI for all of our development projects unless you specifically request us to not use it. If there is any potential issue with data that could be considered confidential (governed by HIPPA or other online data rules), we'll contact you before using AI on your project.
We are now at a point where we can honestly say we don't have to limit projects to small Minimum Viable Product (MVP) scopes when we start. In the same time (and often similar or lower cost) we can create a fully scoped project within the same time frame.
So if you have been thinking about adding some new custom functionality to your web site - even integrations with completely external systems- those things are no longer an option that is too pricey to consider.
Give us a call or click the free consultation button above to set a time to talk about your project and how we can harness AI to make it happen faster and cheaper than ever before.