Of Earth and Stars

Note: this is not a thinkpiece and there is no need to debate it or repost it or comment about it. It offers no conclusions and takes no sides besides the one I've already admitted publicly (a reluctant but fatalistic willingness to use LLMs day-to-day, because they seem to work). It's mostly just a journal entry noting the occurrence of a significant change in the nature of my profession. I've turned off comments as I do usually for "things people are likely to heckle me about pointlessly anyways" because I'm tired and don't have patience for that.

With that out of the way: 2025 (particularly near the end of it) and early 2026 have been, for my corner of the software industry, extremely unusual times.

LLMs turned a corner. I'm not sure how else to put it. If you are not interacting with them yet in your day job, you are perhaps lucky, perhaps unlucky, I'm not sure how to judge that but you are definitely operating in some level of ignorance of what has occurred. You may be seeing the 2nd order effects and hiding. You may be telling yourself nothing's changed and it's all just smoke and mirrors, a marketing campaign by con artists aimed at the gullible. I wish it was. But as far as I can tell this is not so: LLMs really, really turned a corner.

Their capabilities expanded a lot. Coding capability seemed like the first bump (especially around the late fall / early winter: the opus 4.5 / gemini 3 / gpt 5.2 series). But it was quickly clear that the capability also extended to something much worse: vulnerability hunting. They can break software even better than they can write it -- I guess because "you only need to be right sometimes" with vulnerability seeking -- and "breaking" has even more people eager for the new capability.

The change has felt, to me, very sudden and very severe. In a matter of months a lot of people I know personally switched from "playing around seeing what I can do" to "I literally never write code by hand anymore" to "my boss is asking whether I can write 100x more code per day and/or firing me" to "help help my team is under attack by hundreds of new security vulnerabilities and can barely keep up".

I still write some code, but less and less, and more of it is around the margins: touchups, sketches of APIs and data structures, subtle stuff it's easy to be subtly-wrong about, or perhaps LLM-supervisory bits. Because the LLM really does often write the main logic as well as I would at this point, and faster, and more persistently. And also I'm now busy responding to all the damn vulnerabilities. There is an arms race, and I'm now plainly in it.

This is the fastest and most violent change to working conditions and assumptions I've witnessed in my career, including the arrival of the internet and open source and distributed version control and cloud computing and all of that. Nothing else is in the same ballpark.

Software projects have tried to adapt. Some are trying to embrace the tools, some are firmly rejecting them. Some have closed their issue trackers to new submissions which were all slop. Some maintainers have quit, some contributors have been banned, some dependencies have been rolled back or severed, some forks are emerging. A lot of people are re-evaluating (and some rebuilding) their entire software stacks. A lot of people are debating licenses again, with even more fury than they did during the drafting of GPLv3.

Thinkpieces on this event proliferated, many very sour. People wrote about mourning their loss of identity as programmers. People wrote about fear for their loss of jobs. People wrote a lot about their personal disgust with the slop, their fury at the billionaires, their sense that all this is part of of the fascist turn of America. The level of anger in the community of programmers is unlike anything I've ever seen before. People are making lists of who's been infected by the menace and who's still clean. The community is tearing itself apart. Professional and volunteer relationships ended, friendships lost, battle lines drawn.

I'm not writing this to come to any particular conclusion, just to note that it's happened, that it's a set of events that I've experienced as they're happening. This is a journal and sometimes all I can do with it is log events. I don't know how this is going to end, or what to make of it all, I really don't. It's sort of interesting, deeply confusing, sometimes sort of fun, mostly sort of horrifying, sort of miserable. The unit economics of making and breaking software in 2026 are completely different than they were in 2025. More than anything, it's just weird.

This time next year we could all be out of work, or dead from a nuclear war, or even-more-burnt-out from sustained 100x higher velocity of code and vulnerabilities with teams of adversarial LLMs, or .. the whole thing could collapse because maybe, just maybe, it really is "all just a bubble" pushed by VCs on credulous rubes like myself, and it'll vanish like a bad dream. I'm not presently betting on that, but I couldn't have predicted this year, so I'm not going to make any predictions about the next.

I guess I'm sorry to anyone who thinks I'm infected, or facilitating the fascists, or whatever. I'm just trying to adapt. I hope you can see me as a human again someday. I miss the past too. I don't see a way to go back to it, but I'd like it too if there were one.