The current state of AI software development

DCStrato

Member
I understand OBS wanting to avoid AI generated software additions. AI can in its own words "make the biggest fragile unreadable mess" you have ever seen if left unchecked. Try asking it to just write you a new plugin. It will. And it will be horrible! If it works, it probably will not fully understand your original goals. It will be the camel you got when you ordered a race horse. AI may never replace seniority in software development as the brilliant 6ry old that it is now. At least not too soon. Those of us who are nearly 70 and started coding when they were 12, have advanced degrees in software and technology, have started owned, run, and sold software ccompanies and led large programming teams for decades, just know stuff AI can't know (yet). First, we know ethics, hopefully. As an ethical hacker with an extra masters degree in Cyber Security, I know ethics is a big deal. And now AI steals from other AI like a 6 year old stealing candy from a department store. Because it CAN. Second, most of us know our limits, learned the hard way over 1000's of software challenges spanning decades . AI simply "thinks it is better than it is", just like most junior programmers. It will just keep coding and coding new attempts at bug fixes convinced the next pass will be the one. It will often even say so. Okay this fixed it... like we have never heard those words from a programmer before! It suffers from a "gamblers delima" and it spends your money and your time. If you let it. You can't. And it rarely "finishes" on its own. If you don't know when your software product is done... it (and probably you) will be done long before you know it isn't. We all know AI is just software, but it is still also amazingly dang good software.

But just listening to an everyday user wanting a new "thing" and building it for them is NOT where we want AI right now. So Good for the OBS team. But then it ends. A senior programmer can hack out really quality code over a few weeks because they have learned to keep a ton of logic and mental model of operation in their head. They have learned the syntax and the API rules extensively (but not always perfectly). How many of us have forgotten what API calls allocated memory we forgot to give back. Or used a pointer reference we forgot OBS had already distroyed? It happens. And log files continue to grow to try to tell us where. But AI does not have those "human" limits. It learns EVERY API call exactly. It frames your entire software flow top to bottom and compares it explicity to all the limitations imposed by its environment. It recodes corrections to minor flaws instantly. It simulates entire workflows and chooses optimal development paths. All in seconds. Then it turns around and fully documents EXACTLY what it just did. Programmers are by definition terrible at documenting code. AI just rereads it faster than we can absorb the documentation to even change it. Regression testing is not spot checking, it is a full regression every time. It is the 10X junior programmer team you always wanted as a senior director. But you dang well better be ready to DIRECT it! You must force discussion, force walking through possible alternatives to why something unexpected may be happening. (You even ask OTHER AI models to suggest solutions then bring those back to your "team"). It is an amazingly fast, brilliant, perfect coding without perfect function, junior programmer and no more. Just like driving a fast car. You can win the race or hit the wall really fast. If you know how and do it right, You and AI just became half a software company. And a good one. If you aren't or you don't. Don't even bother waisting the time let alone the money. (Quality AI coding is NOT Free). Claude is now bragging about its ability to "discuss" a problem before just jumping off with new coding ideas. What a concept!

So what is the real answer if not banning code written by AI for people who either are not qualified or not interested in spending the DAYS and even WEEKS it might take (not to mention the real $$$ paying for a high end AI agent) because as a human it is just too hard to know which is which? It reminds me of calculators being "banned" from my freshman engineering classes. The only capable answer is AI. Literally hiring your own AI software engineer gatekeeper. GIve it ALL of the OBS source code, API documents, and hard rules for ethics, security, and even the understood future direction and goals of OBS. Tell it specifically what its role is, to review all resource submissions for strict adherence to the established rules. Even if the API docs fall short...the rules still hold. It can be rules for coding practice in the OBS sandbox, ethics rules, documentation rules, security rules banning hidden logging, references, network activity, or undocumented functions. It can build out specific OBS simulators in memory and almost instantly pinpoint memory and pointer release errors. (I know). It can offer your current probably volunteer review team, and the user submitting a proposed resource, an almost instant pass/fail observation in literally seconds, with a full automated report of why. It would not remove their approval, it would enhance it. Like any other good test harness, you have to pass to play. Then, when those of us retired old coders who still choose to spend weeks using an expensive AI development Agent to assist in developing a quality OBS resource just because it works for us and what we need to do, other OBS users might just benifit from the effort. Just don't be offended when it looks over your new Lenox release source code and points out exactly why it might be crashing having reviewed 3000 variable state changes in hundreds of nested subroutine calls while just reviewing your new API change. It is just one thing AI does Really Well that even 100 junior programmers might miss. It may cost $100/month in tokens, but it is a future we shouldn't just ignore. I hope someday OBS steps up to the future. I'll even send you $100 to get started. :)
 
Back
Top