I keep going back and forth on this, so I'm writing it out to settle my own thoughts.
What happened to coders was not nothing. Millions of people spent years, decades, building up a body of work. They posted it openly, shared it, contributed to projects because the culture said that's what you do, that's how we build together. And then companies scraped all of it, fed it into models, and turned around and sold access to a thing that could do a rough approximation of what those people spent their careers learning to do. And a lot of those same people got told their roles were shrinking, their rates were dropping, their skills were becoming optional. The code was taken. The credit was not. The compensation was zero. And now we're supposed to just be excited about the technology.
I don't think I owe those companies my enthusiasm. I don't think I owe them my data, my prompts, my subscription fees, or my quiet participation in an ecosystem that treated a whole profession's output as free raw material. Using their servers, their APIs, their cloud-hosted assistants, means feeding the machine that already ate someone else's livelihood. It means my usage patterns become their training data. It means my money becomes their runway to do it again to the next group. I don't want to be part of that loop. I just don't.
But here's the part I had to sit with for a while. The knowledge is still there. The patterns, the techniques, the solutions to ten thousand common problems, the accumulated understanding of how software works, all of that got learned by these models from real people's real work. And I can't pretend that knowledge doesn't exist now, or that refusing to touch any of it is some kind of moral purity. It would just be cutting myself off from a tool, out of principle, while the principle doesn't actually help the people who were wronged. My suffering in silence doesn't put anything back into their pockets.
So the compromise I've landed on is local models. Run them on my own hardware. No server. No API call. No telemetry. No company seeing what I type, no corporation logging my workflow, no monthly payment going to the people who pulled off the largest uncompensated labor transfer in the history of software. The weights are on my disk. The electricity is mine. The inference happens in a box under my desk.
And yeah, those weights were trained on code that coders wrote, and I don't forget that. But using a local model is more like borrowing a book from a library than renting a service from the company that stole the book. I get access to the collective knowledge. I get the autocomplete, the explanation, the quick draft, the debugging nudge. But I give nothing back to the pipeline. No data. No money. No legitimacy. No signal that says this is fine, keep going, do it to the next industry.
It's not a perfect position. I know that. Someone will say the weights themselves are the stolen thing, and they're not entirely wrong. But I think there's a meaningful difference between using a tool that sits on your own machine and subscribing to the service of the company that built its empire on the taking. One is a quiet, personal use of knowledge that's already in the world. The other is handing your support, your data, and your dollars directly to the people who did the taking, and telling them it worked, do it more.
I can hold both things. I can be grateful for what coders built, angry at what was done with it, and still open a terminal and run a model locally to help me through a stubborn bug on a Tuesday night. Those aren't contradictions. They're just a person trying to use a tool without kneeling to the people who misappropriated it.
I don't support the industry. I just use what it made, on my own terms, in my own house, where nobody is watching, and nobody is getting paid for stealing everyone's stuff.
And I think that's as close to right as I can get.