What’s wrong with this community ?

[wmtech]

If similar was required here for the RM submissions, with no judgement applied, just the disclosure of whether it was created fully with AI or created with assistance with AI, I think that would help with some of the concerns.

This is a bit like requiring developers to disclose whether they used a PHP IDE when developing their software. ;-)

AI is becoming an integral part of software development at an incredible pace. Within a few months, using AI assistance will likely be the norm rather than the exception. Eventually, leaving that checkbox unchecked may become increasingly rare, making this information largely meaningless.

 

[Kevin]

This is a bit like requiring developers to disclose whether they used a PHP IDE when developing their software. ;-)

AI is becoming an integral part of software development at an incredible pace. Within a few months, using AI assistance will likely be the norm rather than the exception. Eventually, leaving that checkbox unchecked may become increasingly rare, making this information largely meaningless.

And yet it is something people are asking to know the answer to.

 

[Digital Doctor]

The XF RM isn't ready for the speed of AI coding.
I think it's time for subcategories for the black hole that is ...

How do new Xenforo admins find anything in there ?

Add-ons [2.x]

xenforo.com

Good addons can't be found is a significant problem.

I suppose there a bigger issues at play a XF anyway.

 

[Digital Doctor]

 

[Anthony Parsons]

If you aren't using AI to your advantage, you are disadvantaged today. The best coders in the world have publicly admitted, AI codes faster and better than they can. Why wouldn't you use it? Pride?

Who says you have to disclose it? How about, f**k off?

You can go to bed, awake, and an AI on a task overnight has run endless cycles across your code for security, performance, etc etc etc, you awake and have a list of issues, or it may stroke your pride and tell you the code is perfect. Humans can't keep up with that.

I put AI across cloned production servers, then let it attack from every angle, and every single time they found vulnerabilities that could be exploited, which are now closed on the production servers.

Tech moves to fast for any person to keep up with, you need AI today in software and tech stacks.

The XF RM isn't ready for the speed of AI coding.

I think it needs a second item menu, and a cron. If marked unmaintained, it exists in its own menu.

 

[ES Dev Team]

AI codes worse than me but in cases where speed is more important than quality, and the task doesn't take longer to describe by hand, i'll use it, read the code, test extensively, and do manual cleanup until it's equivalent to a product i would have hand-coded.

Those latter steps - most people won't do when using AI tools. I consider this a lower quality product than a human created one because it's usually the case. So the AI label can save me a lot of wasted time when researching and comparing softwares.

The only exception for me is if a programmer i know to be good already has done something with AI; that person learned hyper-detail orientation and practiced it for years, so that habit is probably in the end product, and it may be OK to Good quality.

I put AI across cloned production servers, then let it attack from every angle, and every single time they found vulnerabilities that could be exploited, which are now closed on the production servers.

I need to do this on a lot of my systems. What kind of tools are you using to 'red team' your softwares?

 

[vladalincik]

Saying "AI codes better than the best coders in the world" is a huge stretch. It’s great at generating boilerplate fast and finding low-hanging security flaws, but it still hallucinates edge cases and writes bloated code if you don't babysit it. It's a power tool, not a replacement for a solid dev.

 

[Anthony Parsons]

AI codes better than the best coders in the world

Ok... Linus Torvald made that statement, but by all means, what does he know! I did reference that earlier somewhere, when this was raised just like it is again. Anthropic, claude code does 100% of the coding. What do they know too I guess!

 

[vladalincik]

Nobody's saying Claude isn't impressive, I use it daily. But there’s a massive difference between generating functional code and writing optimal, secure, edge-case-proof code. Great tool in the hands of a good dev, but relying on it blindly for 100% of production code is a slippery slope.

 

[Anthony Parsons]

there’s a massive difference between generating functional code and writing optimal, secure, edge-case-proof code.

Disagree today. Last year, sure. A lesser AI model, absolutely. Claude and Codex? Nope. Claude is probably the best, but when combined with codex, they're writing equal code to the best in the world, except they're doing it faster. Optimised, secure, edge proof. They're doing it so fast, it makes the best coders in the world look like monkeys. Claude and codex are doing in hours, what a developer took weeks to do.

Fight it all you want. Don't care. The truth is a kick in the nuts, and when I heard Linus Torvald say the statement he said, if I had been drinking coffee I would have spat it across the screen. He is a realist though. The difference in claude and codex between the beginning of this year and now, vastly different. What those two will be able to do next year... my mind boggles at the thought. It's already somewhere between scary and frightening.

 

[ES Dev Team]

I've been running AI models for my dev shop since early 2025 for privacy, reliability, latency, and cost.
I made this image to show my developers the progression of commercial AI vs what we are running on a 3 GPU setup.

I don't think AI model intelligence is a problem today, however, the processes that people generally chose is where the problem starts, and it's how you end up with slop.

Because commercial AI systems are so slow, most people are using agentic coding processes which have the side effect of removing you from the process. Insights that would naturally steer the outcome don't occur. Flaws in your plan that you'd normally notice during the process are also waiting for you in the end product. Problems with the code are also waiting you at the end of the process, and later, since this way of working discourages both that.

So with these processes you are kicking multiple cans down the road and probably never picking them up because that would make you a 2x developer instead of 10x developer.

Instead of this, we use human in the loop processes with high visibility, inside the human work environment. When the AI model makes changes, those changes are highlighted for easy and fast review. Design problems are immediately noticed. No cans get kicked down the road and we consistently end up with human-grade or better product from this process.

We never produce slop because awareness and participation on the programmer's side is as high as it was without AI. This process is slower but it results in shorter, more relevant to the task, and better code quality.

Maintaining architectural and codebase quality is critical in applications we work on; the negative effects of poor code / design choices compound quickly in a large and business critical app. And that's mainly what we work on!

Agentic processes can shine in areas where the code has low criticality / responsibility / isolated context. In these cases, substituting clanker brain for human brain usually works out OK.

The codebases we work on are similar to large corporate codebases. We watched large corporations hose their codebases with agentic processes and then decide to dial it back. They are gradually starting to adopt our ways. Amazon and Microsoft are good examples of this.

In conclusion, i'd say it's not about how big the artificial intelligence is, but it's how you use it!

 

[Semper]

What was the topic of this thread?

 

[Digital Doctor]

What was the topic of this thread?

Spoiler: tl;dr

nothing.