Have you added a big feature after launch that required changes to multiple parts of the codebase?
I doubt that a codebase almost purely built with AI will survive in the long term.
If there is no developer that understands what is happening, after a while the AI will start running in circles due to a lack of guidance and new features or patches become almost impossible to push out reliably.
AI is good at writing small-ish tools from scratch. Most software development isn't like that, and people claiming it is are disingenuous. That being said, it is also true that if you are doing that kind of software development, AI will be very useful to you in the same way someone writing the code for you would be. It works, yay, you don't fully understand it, boo, but it still works. Maintaining it will be more difficult and error-prone, but it'll actually get you to build something, which is great!
“More than meets the eye, Transformers!” But seriously, current LLMs obviously pass the Turing Test. No one twenty years ago would have not assumed this was AGI/Skynet. I think it’s an apt name that describes what it does pretty well, if not how it does that. Like calling “Internal Combustion Engine” a “Horseless carriage”, it’ll fall out of favour eventually.
Don't worry, all those pitches and pivots to Ai don't actually use anymore "Ai" than they did pre LLMs.
They are doing it for the money and attention. Like a problematic child or domestic pet, it's not all their fault - blame the parents.
Whether or not it actually uses "AI", this garbage is still being forced down everybody's throats.
I used AI to write a web app in 2 months that extensively uses AI — pre-prompting via LLMs and image, sound, and animation generation.
Hardly wrote a single line of code myself.
Just crossed 372 sign ups and 73 paying users.
AI has been extremely valuable to me. Objectively speaking I have already gotten thousands of USD in value out of it.
Just my anecdote.
What's the product / domain?
How do you plan to maintain it?
Have you added a big feature after launch that required changes to multiple parts of the codebase?
I doubt that a codebase almost purely built with AI will survive in the long term. If there is no developer that understands what is happening, after a while the AI will start running in circles due to a lack of guidance and new features or patches become almost impossible to push out reliably.
It's https://gametorch.app/
I added a huge feature, the AI sprite animator, which involved a massive complex diff across much of the codebase and database schema.
That new massive change is now my primary source of revenue.
I understand what is happening.
I appreciate that you are so concerned.
AI is good at writing small-ish tools from scratch. Most software development isn't like that, and people claiming it is are disingenuous. That being said, it is also true that if you are doing that kind of software development, AI will be very useful to you in the same way someone writing the code for you would be. It works, yay, you don't fully understand it, boo, but it still works. Maintaining it will be more difficult and error-prone, but it'll actually get you to build something, which is great!
I disagree.
I've found it's very useful at any scale and you can always understand what's going on.
I wish we called it transformer tech rather than AI. It's very unfortunate
“More than meets the eye, Transformers!” But seriously, current LLMs obviously pass the Turing Test. No one twenty years ago would have not assumed this was AGI/Skynet. I think it’s an apt name that describes what it does pretty well, if not how it does that. Like calling “Internal Combustion Engine” a “Horseless carriage”, it’ll fall out of favour eventually.