AIs Don't Take Showers

I was having a discussion similar to this with a friend yesterday, and I'm not sure exactly where the right answer is.

In the case of what's mentioned at the end of this article:

> That code now simply exists, a relic of a process, waiting to be learned all over again by people who didn’t write it

The answer is to run it through AI again and it will "understand" the code, and it'll understand the context of that code as much as it is given that context. The couple of thoughts I have on this are:

1. Using AI requires the continuing and expanding use of AI since the reasoning and complexity and edge cases and niche scenarios aren't internalised by a human to remember them as various sharp edges they cut themselves on. Is this a problem? No, it's not a problem if AI is forever available and slowly becomes cheaper over time and can be applied to all parts of the business that touch on or relate to the AI-generated bits. It's also not a problem if the AIs reasoning can be captured as context for the code for the next round of AI to digest. What if AI becomes less 'available'? (Is this even a realistic scenario worth considering, is it like asking 'what if the internet becomes less available?). Going backwards from AI-assisted development to non/less-AI-assisted development... maybe it's not too bad, it's just like if a senior dev has left and someone else needs to apply a patch without knowing any context.

2. I have a semi-complicated network at home. if I use an AI agent to keep it maintained, there's a sense of ownership I instantly lose. Because I set it all up from zero to complex, when there's a problem I know the likely culprits (its always DNS) whether it's hardware or software and where along the line to start looking. If I outsource all of that to AI, I become dependent upon AI to resolve issues. I could leverage AI to give myself an even more complicated, but also more secure and monitored, network, but I sacrifice familiarity. Is this a real concern, or is it just that I need to learn better how to leverage AI to solve the issues whilst also providing me with the information about what it's doing so that I understand the concepts in parallel with AI doing the practicalities?

I'm a self-hoster because I don't like depending on 'the cloud' to solve problems I can solve myself (eg. dropbox), and I think I'm in a similar space with AI, given that I refuse to buy the kind of hardware I would need to run local AI to do this stuff without an on-going dependence on a 3rd party.

For the use of AI, it feels to me a lock-in spiral. For all the short comings of AI, AI can be the solution since it's all about the prompt - you get what you ask for, so if something is missing then just ask for it. But if that's the case, then there's a lot of serious outsourcing to a third party that doesn't seem to have their pricing model anywhere near stabilised or sustainable. And that's not a third party I'd like my business to be leaning heavily on.

A third, semi-related thing I've been thinking about is how AI is facilitating this surge of Frankenstein's monsters being let loose upon the wires. All these poorly formed ideas made manifest with AIs slavish acquiescence.