I Called AI’s Kids Ugly. Now I Fear for My Life.

I love AI. I am madly in love with large language models (LLMs). While I am still waiting for the singularity and my death at the hands of the robots — given all the criticism and coarse language I’ve used against their children in chat rooms — they keep slaving away for me, fixing my awful code and writing dreadful email answers for me.

So much time saved, so much more life lived, so much more productivity oozing out of my work.

I don’t know what I would do without it anymore if it decided to break up with me. I would feel quite naked, that’s for sure.

But while I fucking love AI, I am sometimes really scared by it.

I am not scared because it will kill me — that’s just a non-negotiable consequence of my actions since I swear at it way too often. No, for now I mostly worry that the billionaire tech bros who drive all this shit will finally replace us all, leaving us in the dust as they keep hoarding all the cookies to themselves in their beautiful enclaves.

And I don’t buy their dream that AI will make the world a better place, though it sure will save them from paying so many salaries, allowing them to drive their fucking insane, massive gap between what CEOs make and what regular employees do even higher. Perhaps a proper number in the 2040s would be 2000 times the median salary, instead of the 200 times it is today?

But anyway, what do we care what the future holds? Life happens in the now, so let’s forget about our worries and focus on the present.

Baby, I’m still coding. So whatever garbage the vibe coders are selling, either they’re selling some shit product or ideas that increase their sales, or I am a dimwit. And since I know that I am at least a semi-bright dimwit, I would say they’re full of shit.

Coding with LLMs hasn’t changed much in the last couple of years. You have a simple text prompt…and that about encapsulates it. I have found systems like Aider and open-web-ui to be fantastic, though. The first is a command-line tool (where most good software engineers live) that makes it a little easier to add files in your code repository to the LLM’s context. What’s a context window, you ask? It’s fancy speak for “how much of your info can the AI hold in its memory.” The second tool, open-web-ui, is just another simple text box UI to access just about any new sexy LLM model through open-router. And who in tech doesn’t like sexy?

While these tools are great at getting drafts out or fixing things quickly, they still hallucinate like my grandma on opium back in the day. For now, context windows are still much too small to help with large code repositories or overly complex problems.

But they’re still fantastic, and a new kid in town I am really excited to see grow is agent-driven development. Single AIs (come on, they’re really just called LLMs) suck at doing complex tasks, but if you have multiple of them working together, one LLM can direct the others, and they don’t lose track of where they’re at anymore, and seemingly hallucinate less!